ROMANTIC UTOPIA
내용
[subtitle-meta] lang=en auto=True
project on an inventory management model based on corporate demand forecasting using AI.
The team name is Dimension 5. The project
participants are Jo Cheong, Son Young-kyung, Ha Tae-soo, Yoon Jun-seok, and Kwon Yeon-ha. The table of
contents for this report is as
follows: First, the report description
introduces the purpose and necessity of the project. Next, the project sequence
explains the goal of optimizing corporate operations through demand and inventory forecasting. Next, the
procedure for processing formulated data into a format suitable for training through the raw data preprocessing process is outlined. Following this, the visualization data of the models investigated by Hee-won is
explained, and the characteristics and performance of each model are
compared. Finally, the conclusion presents the
practical applicability of the final model and expected future
effects. The goal of this project is to
optimize operations by precisely forecasting the demand and inventory quantities of corporate products.
To achieve this, we aim to derive more accurate forecasts by utilizing a demand forecasting model, an inventory
forecasting model, and an ensemble model.
The dataset
consists of a total of five stages. File 1 is the
basic dataset,
constructed by stopping the original data.
File 2 was improved to be suitable for model training through data normalization and the addition of chemical variables.
File 3
enhanced realism by reflecting seasonality and events.
File 4
strengthened realism by adding noise and reflecting exceptional situations, and the
final File 5
is the finalized dataset for LSTM demand forecasting. The
LSTM-based model takes variables from 30 time points as
input, passes them
through an LSTM layer with four units, and
generates seven outputs. The Adam
Optimizer MS loss function
was used for model training, with a width of 50 and a batch size of 32. This
model is used to forecast time-series data and learns the
long-term dependencies of the time-series data through the LSTM layer.
The model
compares actual sales volume with predicted sales volume on a 7-day basis; while the
overall trend aligns, there is a
prediction error at specific points in time. Since the prediction error
increases during periods of rapid sales volume fluctuation, external variables must be
considered. To improve model performance, it is necessary to
increase prediction accuracy during periods of rapid change and continuously monitor the data using various evaluation metrics.
Inventory
fluctuations are proportional to demand, so the model's prediction
accuracy is generally high; however,
prediction accuracy drops due to a lack of data during periods of rapid inflows and outflows.
The model displays a
comparison between actual and predicted inventory levels, and
prediction
accuracy needs to be improved during periods of rapid fluctuation. It is important to increase prediction accuracy regarding
rapid fluctuations through future data acquisition and model improvements. The
wide value distribution of the month indicates that it is an important variable. The closer it is to
the weekend... Although the predicted value
increases, the impact of the number of weekdays is
not significant or acts in a decreasing direction.
Variables related to anniversaries do not have a major impact on the prediction.
Overall, the two
lines flow similarly, and the
stability of the graph is good. However,
in some sections, the prediction
appears to be over- or under-predicted. This may be because the model
reacts less sensitively to extreme values.
While actual sales fluctuate irregularly, the predicted value
remains at an almost constant level.
In other words, it appears that the model is only predicting the average or
simple value.
Looking at this graph, the
predicted value rises in early January but
increases gradually from the middle onwards. However, towards
the end, it remains almost constant, showing a
trend of stabilization. This
graph demonstrates that the prediction results of the Random Forest model
do not properly reflect the rapid volatility of actual sales volume. The
predicted value appears as a smooth curve close to the average and
fails to properly track trends or sharp drops. This
suggests that Random Forest exhibits limitations in predicting irregular patterns or seismic patterns. This graph shows that the non-predicted value
does not properly reflect the large fluctuations of the actual value and remains at the average level. While
actual
sales volume is widely distributed from 30 to 180, the
predicted value is in a narrow range It is
concentrated, which means the model has not
sufficiently learned volatility such as sharp rises or falls. This
graph shows the feature importance of the Random Forest model. The
most important
variable is inventory level; its importance is
overwhelmingly high at over 0.7,
having the greatest impact on sales volume prediction. Weather,
price, event status, and competitor prices also
appeared as variables affecting demand, but they are likely to have a greater impact when combined rather than individually. This
graph visually displays not only the influence of each variable but also the
direction and magnitude of their impact on the predicted values.
Inventory level, price, and
event status emerged as the most important variables
;
predicted values tended to decrease as inventory levels increased and
increase as the number of events increased. This
confirms that the model is effectively learning the factors that influence actual demand. This
graph shows the changes in
training loss and validation loss during the RF model training process.
Performance improved rapidly as the number of trees increased, and it showed stable convergence
after approximately 20 to 30 trees. Since there was
almost no difference between training loss and validation loss, it
can be confirmed that the model trained stably without overfitting. This
graph shows the
results of predicting 10 days of data using the RF model. The
actual and predicted values
generally show similar patterns, with
upward and downward flows being almost identical. It
follows the trend
well, particularly reflecting periods of rapid change, confirming that the model has effectively learned short-term
trends and pattern sensitivity. This
graph
shows the results of comparing actual and predicted values over 10 days for five samples. In all samples, the blue and
orange lines almost overlap,
confirming that the predicted values closely follow the actual values.
In particular, not only the trend but also the
locations of the highs and lows are nearly identical,
indicating the model's excellent pattern learning performance. Although the
Random Forest model
shows a flow similar to the actual values, the
fluctuation range of the predicted values
appears larger than the actual values in some sections. This
phenomenon occurs because Random Forest tends to predict extreme values in certain feature combinations. The overall demand
pattern has been learned well. Some predicted values
exhibit characteristics of being excessively high or low. These
are the prediction results for the data. Since the
actual and predicted values match almost perfectly, it
accurately predicts upward and downward patterns.
In particular, it
demonstrates high accuracy in both rapid rise and stable maintenance periods, confirming that
the model has learned the patterns well. The
sales volume prediction results for 10 days from Sample 0 to Sample 4 have been visualized
in a single graph. The
actual and predicted values
appear mostly similar, and in particular, the upward and
downward patterns match almost perfectly. This
demonstrates the excellent short-term forecasting performance of the model.
The model exhibits consistent performance across various samples and
accurately reflects changes over time.
Comparing the average of the predicted and actual values across the entire sample, the
RF model demonstrated accurate and stable performance not only
in individual predictions but also in the overall average. It
confirms that the model possesses high average reliability, as predictions maintain a consistent scale without consistently underestimating or overestimating within specific ranges.
These results indicate that the RF model
demonstrates overall superior performance in demand forecasting. This
graph shows the results of a
scenario analysis of sales volume changes over the next 10 days, varying prices and inventory levels.
Predicted sales volume
tends to increase as prices drop and inventory levels rise; in particular, the highest sales volume is recorded when the price is
15,000 won and inventory is 150 units.
Conversely,
sales volume plummets when the price is 35,000 won, confirming that demand elasticity plays a significant role. This
graph visualizes the importance of permutation characteristics.
Event presence and price
emerged as the most important characteristics, while competitor
pricing and lead time also showed relatively high
contributions. On the other hand, seasonal variables
showed negative importance, suggesting that the model
deemed seasonality unimportant.
In other words, the most
important variable in sales volume forecasting is It can be seen that these factors, such as events and prices,
have a significant impact on sales. Factors
such as lead time, event availability, and price
have a significant influence on model forecasting; in particular, lead
time is
important for inventory management and delivery strategies as it represents the supply preparation period.
Seasonal variables can actually degrade model performance, so they
should be removed or handled in a different way.
Variables showing low importance, such as holidays, weekend availability, and weather,
may be significant within specific categories, so
granular analysis is required.
The variables that have the greatest impact on sales volume forecasting are inventory quantity, price, and competitor price, and
these have a positive correlation with sales volume. Factors such as
season, holidays, and weekend availability have
low influence based on linear correlation, but they can interact in non-linear models, so their removal
must be done with caution.
To improve model performance, non-linear relationships must be
considered, relationships between variables analyzed using various models, and domain
knowledge
utilized. There is no distinct linear relationship between mid-time and sales volume, and the
correlation between the two variables is
very low. Some data points
deviate from the overall distribution
and may be considered outliers, so
caution is required during data analysis. Since factors other than lead time
can affect sales volume,
additional analysis and data
collection are necessary. The TFT model shows
nearly equal importance to the workforce variable at the last encoder point. It assigns weights, meaning the
model makes judgments by combining various variables rather than relying excessively on specific variables.
The weights of all variables
are similar at a level of approximately 0.1, showing a pattern different from the general assumption that
high selection weights are assigned to specific variables.
Data normalization or attention layer
tuning may still be in the early stages; while the
model's generalization performance is high,
model improvement may be considered as needed.
Variable Selection Ways
assign equal importance to all variables, whereas
Permutation
Forces show a significant difference in importance per variable.
Variable Selection Ways
indicate that the model predicts by combining various variables, while Permutation
Forces indicate that the model relies on specific variables.
When developing a model, both
methods should be used together to
comprehensively assess variable importance and improve model performance. This
graph
visually
displays model performance by comparing the prediction results of various models, such as Random Forest, TFT, and Ensemble, with actual values. By comparing
the predicted values of each model with the actual
values over 50 data points, you
can identify which model provides the most accurate prediction.
The graph helps
analyze errors by model and assists in selecting the most suitable model.
Overall, Ensemble models
predict sales volume lower than actual sales volume. It
exhibits a predictive tendency, with a
significant prediction error
appearing particularly on days 4, 7, and 8. Problems arise where prediction sensitivity is insufficient in high-demand periods, and
prediction accuracy drops in periods of large fluctuations, such as immediately after marketing events or restocking.
Particular caution is required in periods where demand forecasting failure can have a significant impact on the business, and
separate
modeling strategies for high volatility should be considered alongside model improvements. The TFT
model is a deep learning-based model that predicts future demand by learning from time-series data. It
has a structure capable of effectively forecasting demand by considering multivariate characteristics.
In this study, the model was trained using data from the past 30 days
as the encoder and a forecast period of 7 days. This
graph compares the
prediction results of the TFT model and the RF model
with actual values.
By
visualizing the 30 samples prior to the day 7 prediction value, it shows how closely each model matches the actual values.
By comparing the prediction patterns of the two models, the
performance superiority of the TFT model can be visually evaluated. This
table numerically represents the performance of the TFT model.
Through each value, the model's prediction error
can be intuitively verified. The TFT
model's error is
29.83%, suggesting potential for improving prediction accuracy.
The graph
shows the change in MSE according to the prediction order. As the prediction order increases, the
MSE
shows a pattern of gradually increasing, reflecting the
general characteristic of time series forecasting where errors increase during future predictions. This
graph
compares the average predicted value of the entire sample with the actual value. It allows for
the identification of the model's overall trend and verification of how well the model tracks the variability of the actual value.
Based on this graph, it can be concluded that
training the dataset using an ensemble model is better than coding TFT alone. This
graph compares
the predicted values of five randomly selected samples with the actual
values. It allows for the evaluation of the model's short-term
prediction performance and shows
how well the model has learned specific patterns for each sample. The
LSTM model
is a deep learning model suitable for predicting the future by learning from time series data. In
this study, we utilized LSTM to
predict future inventory levels based on corporate inventory data. This graph compares the
inventory levels predicted by the LSTM model with the actual inventory levels. It allows for an intuitive verification of
how well the LSTM model predicts inventory variability over time. This
graph visually demonstrates the performance by
comparing the predicted values of the XGB ensemble model with the actual values.
How the predicted values differ from the actual values You
can evaluate the accuracy of the model by checking for similarity. The
more similar the predicted value is to the actual value, the
better the model performance. This graph
visually displays the performance of the XGB modal BL model by comparing its
predicted values with the actual values. You can evaluate the
accuracy of the model by checking how similar the predicted value is to the actual value. The
more similar the predicted value is to the actual value, the better the model performance. This
graph
visualizes feature importance based on the XGB model and
shows the impact each feature has on the model's prediction. For example, the
characteristics of cod and price were found to have the greatest impact on the model's prediction.
This allows you to
visually confirm which characteristics the model considers important. This
graph visualizes the feature importance of the ensemble model. The
horizontal axis represents feature importance, and
the longer the bar, the greater the impact the feature has on the model's prediction.
Inventory
was found to be the most important characteristic,
followed by weather and price, which
were analyzed to have a significant impact. Through this, you can see
which characteristics the ensemble model focuses on to perform predictions. This
graph shows the changes in training and validation losses of the ensemble model. Although the
initial loss
value is somewhat high, you can observe a
trend where the loss steadily decreases as training progresses.
In particular, the training loss and
validation loss are similar. As it is reduced into a pattern, it
indicates that the model is stably learning the data patterns.
Through this, it can be concluded that the ensemble model is learning while maintaining general performance without overfitting. This
graph
visualizes demand fluctuations based on holidays and price changes using a decoder. It
intuitively shows how demand varies by price range and various holidays such as Children's Day and Chuseok.
This
provides useful data for establishing marketing strategies. This
graph evaluates model performance by
comparing the actual and predicted values of five randomly selected samples. It allows you to verify
how well the model has learned specific patterns. This
graph demonstrates the efficiency of inventory management by visualizing optimal
inventory levels and predicted demand.
Considering safety stock levels, an appropriate inventory level capable of responding to demand fluctuations
was set, and the estimated safety stock level
is 34.02 units. The analysis results showed that the
TFT model and the xgb positive model
demonstrated excellent performance; in particular, the xgb
non-positive model
recorded the lowest error rate of 17.7%. In the future, we plan to further improve model performance by addressing the
issue of resistance to learning speed due to increased data volume and
strengthening the reflection of external variables. This
demand forecasting model can be effectively utilized for corporate inventory management and the
establishment of marketing strategies. This
chart shows the artificial neural network
model learning It demonstrates that the loss value was stably reduced during the process,
suggesting that the model can produce good predictive performance on the given data. This is a chart visualizing the predicted demand using an artificial neural network, the
safety stock, and the
optimal stock level calculated by summing them. The
optimal stock level
varies significantly depending on fluctuations in predicted demand; while
predicted demand
shows a relatively constant pattern, it exhibits large fluctuations in some sections. The
safety stock
level remains stable, allowing the company to
prepare for sudden demand fluctuations or supply disruptions.
Visualization of training history using SVM demonstrates high prediction accuracy
and generalization. The actual and predicted values are
distributed close to the x-line, demonstrating the model's high predictive performance and
excellent generalization ability.
No overfitting: As the predictive performance of the training and test data is
similar, no signs of overfitting are
present. Presence of spectra and
potential for model improvement: Predictive spectra exist, and model performance can be further improved through
spectra distribution analysis and hyperparameter
tuning.
All new models show a
similarly relatively constant pattern, but exhibit
large fluctuations in some sections. The
XGB model demonstrates excellent predictive
performance and fast training speed and
can be applied to various types of data.
However, there is a risk of overfitting, and
hyperparameter tuning is required.
The model leverages its strength in learning complex non-linear relationships.
Although visible, hyperparameter
tuning is crucial and
there is a risk of overfitting. SVM models
demonstrate robust performance in non-linear regression problems and
can achieve good results on relatively small datasets.
However, hyperparameter tuning is
critical, and training times may be long when the data size is large.
In this project, we compared and analyzed various models with the goal of operational optimization by precisely forecasting the demand and inventory of corporate products. As a
result, the
ensemble model combining TFT, LSTM, and XGB
demonstrated the best performance and recorded high
prediction accuracy.
Through this, we derived meaningful results confirming the
potential for practical application in demand forecasting and inventory management.
Thank you.
원문 자막 펼치기
WEBVTT Kind: captions Language: en 00:00:02.389 --> 00:00:02.399 align:start position:0% 00:00:02.399 --> 00:00:04.670 align:start position:0% project <00:00:02.482><c>on </c><00:00:02.565><c>an </c><00:00:02.648><c>inventory </c><00:00:02.731><c>management </c><00:00:02.814><c>model </c><00:00:02.897><c>based </c><00:00:02.980><c>on </c><00:00:03.063><c>corporate </c><00:00:03.146><c>demand </c><00:00:03.229><c>forecasting </c><00:00:03.312><c>using </c><00:00:03.395><c>AI.</c> 00:00:04.670 --> 00:00:04.680 align:start position:0% project on an inventory management model based on corporate demand forecasting using AI. 00:00:04.680 --> 00:00:07.950 align:start position:0% project on an inventory management model based on corporate demand forecasting using AI. The <00:00:05.062><c>team </c><00:00:05.444><c>name </c><00:00:05.826><c>is </c><00:00:06.208><c>Dimension </c><00:00:06.590><c>5. </c><00:00:06.972><c>The </c><00:00:07.354><c>project</c> 00:00:07.950 --> 00:00:07.960 align:start position:0% The team name is Dimension 5. The project 00:00:07.960 --> 00:00:12.230 align:start position:0% The team name is Dimension 5. The project participants <00:00:08.186><c>are </c><00:00:08.412><c>Jo </c><00:00:08.638><c>Cheong, </c><00:00:08.864><c>Son </c><00:00:09.090><c>Young-kyung, </c><00:00:09.316><c>Ha </c><00:00:09.542><c>Tae-soo, </c><00:00:09.768><c>Yoon </c><00:00:09.994><c>Jun-seok, </c><00:00:10.220><c>and </c><00:00:10.446><c>Kwon </c><00:00:10.672><c>Yeon-ha. </c><00:00:10.898><c>The </c><00:00:11.124><c>table </c><00:00:11.350><c>of</c> 00:00:12.230 --> 00:00:12.240 align:start position:0% participants are Jo Cheong, Son Young-kyung, Ha Tae-soo, Yoon Jun-seok, and Kwon Yeon-ha. The table of 00:00:12.240 --> 00:00:15.470 align:start position:0% participants are Jo Cheong, Son Young-kyung, Ha Tae-soo, Yoon Jun-seok, and Kwon Yeon-ha. The table of contents <00:00:12.695><c>for </c><00:00:13.150><c>this </c><00:00:13.605><c>report </c><00:00:14.060><c>is </c><00:00:14.515><c>as</c> 00:00:15.470 --> 00:00:15.480 align:start position:0% contents for this report is as 00:00:15.480 --> 00:00:18.790 align:start position:0% contents for this report is as follows: <00:00:16.199><c>First, </c><00:00:16.918><c>the </c><00:00:17.637><c>report </c><00:00:18.356><c>description</c> 00:00:18.790 --> 00:00:22.349 align:start position:0% follows: First, the report description 00:00:22.349 --> 00:00:22.359 align:start position:0% 00:00:22.359 --> 00:00:24.390 align:start position:0% introduces <00:00:22.508><c>the </c><00:00:22.657><c>purpose </c><00:00:22.806><c>and </c><00:00:22.955><c>necessity </c><00:00:23.104><c>of </c><00:00:23.253><c>the </c><00:00:23.402><c>project. </c><00:00:23.551><c>Next, </c><00:00:23.700><c>the </c><00:00:23.849><c>project </c><00:00:23.998><c>sequence</c> 00:00:24.390 --> 00:00:26.710 align:start position:0% introduces the purpose and necessity of the project. Next, the project sequence 00:00:26.710 --> 00:00:26.720 align:start position:0% 00:00:26.720 --> 00:00:29.470 align:start position:0% explains <00:00:26.855><c>the </c><00:00:26.990><c>goal </c><00:00:27.125><c>of </c><00:00:27.260><c>optimizing </c><00:00:27.395><c>corporate </c><00:00:27.530><c>operations </c><00:00:27.665><c>through </c><00:00:27.800><c>demand </c><00:00:27.935><c>and </c><00:00:28.070><c>inventory </c><00:00:28.205><c>forecasting. </c><00:00:28.340><c>Next, </c><00:00:28.475><c>the</c> 00:00:29.470 --> 00:00:31.630 align:start position:0% explains the goal of optimizing corporate operations through demand and inventory forecasting. Next, the 00:00:31.630 --> 00:00:34.990 align:start position:0% 00:00:34.990 --> 00:00:35.000 align:start position:0% 00:00:35.000 --> 00:00:38.430 align:start position:0% procedure <00:00:35.098><c>for </c><00:00:35.196><c>processing </c><00:00:35.294><c>formulated </c><00:00:35.392><c>data </c><00:00:35.490><c>into </c><00:00:35.588><c>a </c><00:00:35.686><c>format </c><00:00:35.784><c>suitable </c><00:00:35.882><c>for </c><00:00:35.980><c>training </c><00:00:36.078><c>through </c><00:00:36.176><c>the </c><00:00:36.274><c>raw </c><00:00:36.372><c>data </c><00:00:36.470><c>preprocessing </c><00:00:36.568><c>process </c><00:00:36.666><c>is </c><00:00:36.764><c>outlined. </c><00:00:36.862><c>Following </c><00:00:36.960><c>this, </c><00:00:37.058><c>the </c><00:00:37.156><c>visualization </c><00:00:37.254><c>data </c><00:00:37.352><c>of </c><00:00:37.450><c>the </c><00:00:37.548><c>models </c><00:00:37.646><c>investigated </c><00:00:37.744><c>by </c><00:00:37.842><c>Hee-won </c><00:00:37.940><c>is</c> 00:00:38.430 --> 00:00:40.910 align:start position:0% procedure for processing formulated data into a format suitable for training through the raw data preprocessing process is outlined. Following this, the visualization data of the models investigated by Hee-won is 00:00:40.910 --> 00:00:40.920 align:start position:0% 00:00:40.920 --> 00:00:44.029 align:start position:0% explained, <00:00:41.173><c>and </c><00:00:41.426><c>the </c><00:00:41.679><c>characteristics </c><00:00:41.932><c>and </c><00:00:42.185><c>performance </c><00:00:42.438><c>of </c><00:00:42.691><c>each </c><00:00:42.944><c>model </c><00:00:43.197><c>are</c> 00:00:44.029 --> 00:00:44.039 align:start position:0% explained, and the characteristics and performance of each model are 00:00:44.039 --> 00:00:46.869 align:start position:0% explained, and the characteristics and performance of each model are compared. <00:00:44.551><c>Finally, </c><00:00:45.063><c>the </c><00:00:45.575><c>conclusion </c><00:00:46.087><c>presents </c><00:00:46.599><c>the</c> 00:00:46.869 --> 00:00:46.879 align:start position:0% compared. Finally, the conclusion presents the 00:00:46.879 --> 00:00:49.670 align:start position:0% compared. Finally, the conclusion presents the practical <00:00:47.194><c>applicability </c><00:00:47.509><c>of </c><00:00:47.824><c>the </c><00:00:48.139><c>final </c><00:00:48.454><c>model </c><00:00:48.769><c>and </c><00:00:49.084><c>expected </c><00:00:49.399><c>future</c> 00:00:49.670 --> 00:00:49.680 align:start position:0% practical applicability of the final model and expected future 00:00:49.680 --> 00:00:52.950 align:start position:0% practical applicability of the final model and expected future effects. <00:00:50.017><c>The </c><00:00:50.354><c>goal </c><00:00:50.691><c>of </c><00:00:51.028><c>this </c><00:00:51.365><c>project </c><00:00:51.702><c>is </c><00:00:52.039><c>to</c> 00:00:52.950 --> 00:00:55.310 align:start position:0% effects. The goal of this project is to 00:00:55.310 --> 00:00:55.320 align:start position:0% 00:00:55.320 --> 00:00:58.549 align:start position:0% optimize <00:00:55.503><c>operations </c><00:00:55.686><c>by </c><00:00:55.869><c>precisely </c><00:00:56.052><c>forecasting </c><00:00:56.235><c>the </c><00:00:56.418><c>demand </c><00:00:56.601><c>and </c><00:00:56.784><c>inventory </c><00:00:56.967><c>quantities </c><00:00:57.150><c>of </c><00:00:57.333><c>corporate </c><00:00:57.516><c>products.</c> 00:00:58.549 --> 00:00:58.559 align:start position:0% optimize operations by precisely forecasting the demand and inventory quantities of corporate products. 00:00:58.559 --> 00:01:01.910 align:start position:0% optimize operations by precisely forecasting the demand and inventory quantities of corporate products. To <00:00:58.704><c>achieve </c><00:00:58.849><c>this, </c><00:00:58.994><c>we </c><00:00:59.139><c>aim </c><00:00:59.284><c>to </c><00:00:59.429><c>derive </c><00:00:59.574><c>more </c><00:00:59.719><c>accurate </c><00:00:59.864><c>forecasts </c><00:01:00.009><c>by </c><00:01:00.154><c>utilizing </c><00:01:00.299><c>a </c><00:01:00.444><c>demand </c><00:01:00.589><c>forecasting </c><00:01:00.734><c>model, </c><00:01:00.879><c>an </c><00:01:01.024><c>inventory</c> 00:01:01.910 --> 00:01:01.920 align:start position:0% To achieve this, we aim to derive more accurate forecasts by utilizing a demand forecasting model, an inventory 00:01:01.920 --> 00:01:04.070 align:start position:0% To achieve this, we aim to derive more accurate forecasts by utilizing a demand forecasting model, an inventory forecasting <00:01:02.272><c>model, </c><00:01:02.624><c>and </c><00:01:02.976><c>an </c><00:01:03.328><c>ensemble </c><00:01:03.680><c>model.</c> 00:01:04.070 --> 00:01:06.149 align:start position:0% forecasting model, and an ensemble model. 00:01:06.149 --> 00:01:06.159 align:start position:0% 00:01:06.159 --> 00:01:09.270 align:start position:0% The <00:01:09.040><c>dataset</c> 00:01:09.270 --> 00:01:09.280 align:start position:0% The dataset 00:01:09.280 --> 00:01:12.550 align:start position:0% The dataset consists <00:01:09.531><c>of </c><00:01:09.782><c>a </c><00:01:10.033><c>total </c><00:01:10.284><c>of </c><00:01:10.535><c>five </c><00:01:10.786><c>stages. </c><00:01:11.037><c>File </c><00:01:11.288><c>1 </c><00:01:11.539><c>is </c><00:01:11.790><c>the</c> 00:01:12.550 --> 00:01:12.560 align:start position:0% consists of a total of five stages. File 1 is the 00:01:12.560 --> 00:01:15.070 align:start position:0% consists of a total of five stages. File 1 is the basic <00:01:14.520><c>dataset,</c> 00:01:15.070 --> 00:01:15.080 align:start position:0% basic dataset, 00:01:15.080 --> 00:01:17.429 align:start position:0% basic dataset, constructed <00:01:15.464><c>by </c><00:01:15.848><c>stopping </c><00:01:16.232><c>the </c><00:01:16.616><c>original </c><00:01:17.000><c>data.</c> 00:01:17.429 --> 00:01:17.439 align:start position:0% constructed by stopping the original data. 00:01:17.439 --> 00:01:19.789 align:start position:0% constructed by stopping the original data. File <00:01:17.554><c>2 </c><00:01:17.669><c>was </c><00:01:17.784><c>improved </c><00:01:17.899><c>to </c><00:01:18.014><c>be </c><00:01:18.129><c>suitable </c><00:01:18.244><c>for </c><00:01:18.359><c>model </c><00:01:18.474><c>training </c><00:01:18.589><c>through </c><00:01:18.704><c>data </c><00:01:18.819><c>normalization </c><00:01:18.934><c>and </c><00:01:19.049><c>the </c><00:01:19.164><c>addition </c><00:01:19.279><c>of </c><00:01:19.394><c>chemical </c><00:01:19.509><c>variables.</c> 00:01:19.789 --> 00:01:22.149 align:start position:0% File 2 was improved to be suitable for model training through data normalization and the addition of chemical variables. 00:01:22.149 --> 00:01:22.159 align:start position:0% 00:01:22.159 --> 00:01:25.550 align:start position:0% File <00:01:24.720><c>3</c> 00:01:25.550 --> 00:01:25.560 align:start position:0% File 3 00:01:25.560 --> 00:01:28.910 align:start position:0% File 3 enhanced <00:01:25.953><c>realism </c><00:01:26.346><c>by </c><00:01:26.739><c>reflecting </c><00:01:27.132><c>seasonality </c><00:01:27.525><c>and </c><00:01:27.918><c>events.</c> 00:01:28.910 --> 00:01:28.920 align:start position:0% enhanced realism by reflecting seasonality and events. 00:01:28.920 --> 00:01:31.030 align:start position:0% enhanced realism by reflecting seasonality and events. File <00:01:30.759><c>4</c> 00:01:31.030 --> 00:01:31.040 align:start position:0% File 4 00:01:31.040 --> 00:01:33.749 align:start position:0% File 4 strengthened <00:01:31.224><c>realism </c><00:01:31.408><c>by </c><00:01:31.592><c>adding </c><00:01:31.776><c>noise </c><00:01:31.960><c>and </c><00:01:32.144><c>reflecting </c><00:01:32.328><c>exceptional </c><00:01:32.512><c>situations, </c><00:01:32.696><c>and </c><00:01:32.880><c>the</c> 00:01:33.749 --> 00:01:33.759 align:start position:0% strengthened realism by adding noise and reflecting exceptional situations, and the 00:01:33.759 --> 00:01:36.550 align:start position:0% strengthened realism by adding noise and reflecting exceptional situations, and the final <00:01:34.999><c>File </c><00:01:36.239><c>5</c> 00:01:36.550 --> 00:01:36.560 align:start position:0% final File 5 00:01:36.560 --> 00:01:38.670 align:start position:0% final File 5 is <00:01:36.725><c>the </c><00:01:36.890><c>finalized </c><00:01:37.055><c>dataset </c><00:01:37.220><c>for </c><00:01:37.385><c>LSTM </c><00:01:37.550><c>demand </c><00:01:37.715><c>forecasting. </c><00:01:37.880><c>The</c> 00:01:38.670 --> 00:01:38.680 align:start position:0% is the finalized dataset for LSTM demand forecasting. The 00:01:38.680 --> 00:01:42.149 align:start position:0% is the finalized dataset for LSTM demand forecasting. The LSTM-based <00:01:39.055><c>model </c><00:01:39.430><c>takes </c><00:01:39.805><c>variables </c><00:01:40.180><c>from </c><00:01:40.555><c>30 </c><00:01:40.930><c>time </c><00:01:41.305><c>points </c><00:01:41.680><c>as</c> 00:01:42.149 --> 00:01:42.159 align:start position:0% LSTM-based model takes variables from 30 time points as 00:01:42.159 --> 00:01:44.510 align:start position:0% LSTM-based model takes variables from 30 time points as input, <00:01:43.119><c>passes </c><00:01:44.079><c>them</c> 00:01:44.510 --> 00:01:44.520 align:start position:0% input, passes them 00:01:44.520 --> 00:01:47.350 align:start position:0% input, passes them through <00:01:44.839><c>an </c><00:01:45.158><c>LSTM </c><00:01:45.477><c>layer </c><00:01:45.796><c>with </c><00:01:46.115><c>four </c><00:01:46.434><c>units, </c><00:01:46.753><c>and</c> 00:01:47.350 --> 00:01:47.360 align:start position:0% through an LSTM layer with four units, and 00:01:47.360 --> 00:01:50.709 align:start position:0% through an LSTM layer with four units, and generates <00:01:48.080><c>seven </c><00:01:48.800><c>outputs. </c><00:01:49.520><c>The </c><00:01:50.240><c>Adam</c> 00:01:50.709 --> 00:01:50.719 align:start position:0% generates seven outputs. The Adam 00:01:50.719 --> 00:01:53.149 align:start position:0% generates seven outputs. The Adam Optimizer <00:01:51.385><c>MS </c><00:01:52.051><c>loss </c><00:01:52.717><c>function</c> 00:01:53.149 --> 00:01:53.159 align:start position:0% Optimizer MS loss function 00:01:53.159 --> 00:01:56.230 align:start position:0% Optimizer MS loss function was <00:01:53.329><c>used </c><00:01:53.499><c>for </c><00:01:53.669><c>model </c><00:01:53.839><c>training, </c><00:01:54.009><c>with </c><00:01:54.179><c>a </c><00:01:54.349><c>width </c><00:01:54.519><c>of </c><00:01:54.689><c>50 </c><00:01:54.859><c>and </c><00:01:55.029><c>a </c><00:01:55.199><c>batch </c><00:01:55.369><c>size </c><00:01:55.539><c>of </c><00:01:55.709><c>32. </c><00:01:55.879><c>This</c> 00:01:56.230 --> 00:01:56.240 align:start position:0% was used for model training, with a width of 50 and a batch size of 32. This 00:01:56.240 --> 00:01:57.389 align:start position:0% was used for model training, with a width of 50 and a batch size of 32. This 00:01:57.389 --> 00:01:57.399 align:start position:0% 00:01:57.399 --> 00:02:00.310 align:start position:0% model <00:01:57.674><c>is </c><00:01:57.949><c>used </c><00:01:58.224><c>to </c><00:01:58.499><c>forecast </c><00:01:58.774><c>time-series </c><00:01:59.049><c>data </c><00:01:59.324><c>and </c><00:01:59.599><c>learns </c><00:01:59.874><c>the</c> 00:02:00.310 --> 00:02:02.870 align:start position:0% model is used to forecast time-series data and learns the 00:02:02.870 --> 00:02:02.880 align:start position:0% 00:02:02.880 --> 00:02:05.590 align:start position:0% long-term <00:02:03.115><c>dependencies </c><00:02:03.350><c>of </c><00:02:03.585><c>the </c><00:02:03.820><c>time-series </c><00:02:04.055><c>data </c><00:02:04.290><c>through </c><00:02:04.525><c>the </c><00:02:04.760><c>LSTM </c><00:02:04.995><c>layer.</c> 00:02:05.590 --> 00:02:05.600 align:start position:0% long-term dependencies of the time-series data through the LSTM layer. 00:02:05.600 --> 00:02:08.869 align:start position:0% long-term dependencies of the time-series data through the LSTM layer. The <00:02:08.399><c>model</c> 00:02:08.869 --> 00:02:08.879 align:start position:0% The model 00:02:08.879 --> 00:02:11.550 align:start position:0% The model compares <00:02:09.029><c>actual </c><00:02:09.179><c>sales </c><00:02:09.329><c>volume </c><00:02:09.479><c>with </c><00:02:09.629><c>predicted </c><00:02:09.779><c>sales </c><00:02:09.929><c>volume </c><00:02:10.079><c>on </c><00:02:10.229><c>a </c><00:02:10.379><c>7-day </c><00:02:10.529><c>basis; </c><00:02:10.679><c>while </c><00:02:10.829><c>the</c> 00:02:11.550 --> 00:02:11.560 align:start position:0% compares actual sales volume with predicted sales volume on a 7-day basis; while the 00:02:11.560 --> 00:02:14.070 align:start position:0% compares actual sales volume with predicted sales volume on a 7-day basis; while the overall <00:02:11.984><c>trend </c><00:02:12.408><c>aligns, </c><00:02:12.832><c>there </c><00:02:13.256><c>is </c><00:02:13.680><c>a</c> 00:02:14.070 --> 00:02:14.080 align:start position:0% overall trend aligns, there is a 00:02:14.080 --> 00:02:16.949 align:start position:0% overall trend aligns, there is a prediction <00:02:14.267><c>error </c><00:02:14.454><c>at </c><00:02:14.641><c>specific </c><00:02:14.828><c>points </c><00:02:15.015><c>in </c><00:02:15.202><c>time. </c><00:02:15.389><c>Since </c><00:02:15.576><c>the </c><00:02:15.763><c>prediction </c><00:02:15.950><c>error</c> 00:02:16.949 --> 00:02:19.710 align:start position:0% prediction error at specific points in time. Since the prediction error 00:02:19.710 --> 00:02:19.720 align:start position:0% 00:02:19.720 --> 00:02:21.949 align:start position:0% increases <00:02:19.880><c>during </c><00:02:20.040><c>periods </c><00:02:20.200><c>of </c><00:02:20.360><c>rapid </c><00:02:20.520><c>sales </c><00:02:20.680><c>volume </c><00:02:20.840><c>fluctuation, </c><00:02:21.000><c>external </c><00:02:21.160><c>variables </c><00:02:21.320><c>must </c><00:02:21.480><c>be</c> 00:02:21.949 --> 00:02:21.959 align:start position:0% increases during periods of rapid sales volume fluctuation, external variables must be 00:02:21.959 --> 00:02:24.470 align:start position:0% increases during periods of rapid sales volume fluctuation, external variables must be considered. <00:02:22.224><c>To </c><00:02:22.489><c>improve </c><00:02:22.754><c>model </c><00:02:23.019><c>performance, </c><00:02:23.284><c>it </c><00:02:23.549><c>is </c><00:02:23.814><c>necessary </c><00:02:24.079><c>to</c> 00:02:24.470 --> 00:02:27.070 align:start position:0% considered. To improve model performance, it is necessary to 00:02:27.070 --> 00:02:27.080 align:start position:0% 00:02:27.080 --> 00:02:29.990 align:start position:0% increase <00:02:27.192><c>prediction </c><00:02:27.304><c>accuracy </c><00:02:27.416><c>during </c><00:02:27.528><c>periods </c><00:02:27.640><c>of </c><00:02:27.752><c>rapid </c><00:02:27.864><c>change </c><00:02:27.976><c>and </c><00:02:28.088><c>continuously </c><00:02:28.200><c>monitor </c><00:02:28.312><c>the </c><00:02:28.424><c>data </c><00:02:28.536><c>using </c><00:02:28.648><c>various </c><00:02:28.760><c>evaluation </c><00:02:28.872><c>metrics.</c> 00:02:29.990 --> 00:02:30.000 align:start position:0% increase prediction accuracy during periods of rapid change and continuously monitor the data using various evaluation metrics. 00:02:30.000 --> 00:02:32.750 align:start position:0% increase prediction accuracy during periods of rapid change and continuously monitor the data using various evaluation metrics. Inventory 00:02:32.750 --> 00:02:32.760 align:start position:0% Inventory 00:02:32.760 --> 00:02:35.229 align:start position:0% Inventory fluctuations <00:02:33.030><c>are </c><00:02:33.300><c>proportional </c><00:02:33.570><c>to </c><00:02:33.840><c>demand, </c><00:02:34.110><c>so </c><00:02:34.380><c>the </c><00:02:34.650><c>model's </c><00:02:34.920><c>prediction</c> 00:02:35.229 --> 00:02:35.239 align:start position:0% fluctuations are proportional to demand, so the model's prediction 00:02:35.239 --> 00:02:37.630 align:start position:0% fluctuations are proportional to demand, so the model's prediction accuracy <00:02:35.719><c>is </c><00:02:36.199><c>generally </c><00:02:36.679><c>high; </c><00:02:37.159><c>however,</c> 00:02:37.630 --> 00:02:37.640 align:start position:0% accuracy is generally high; however, 00:02:37.640 --> 00:02:40.190 align:start position:0% accuracy is generally high; however, prediction <00:02:37.784><c>accuracy </c><00:02:37.928><c>drops </c><00:02:38.072><c>due </c><00:02:38.216><c>to </c><00:02:38.360><c>a </c><00:02:38.504><c>lack </c><00:02:38.648><c>of </c><00:02:38.792><c>data </c><00:02:38.936><c>during </c><00:02:39.080><c>periods </c><00:02:39.224><c>of </c><00:02:39.368><c>rapid </c><00:02:39.512><c>inflows </c><00:02:39.656><c>and </c><00:02:39.800><c>outflows.</c> 00:02:40.190 --> 00:02:40.200 align:start position:0% prediction accuracy drops due to a lack of data during periods of rapid inflows and outflows. 00:02:40.200 --> 00:02:43.190 align:start position:0% prediction accuracy drops due to a lack of data during periods of rapid inflows and outflows. The <00:02:41.040><c>model </c><00:02:41.880><c>displays </c><00:02:42.720><c>a</c> 00:02:43.190 --> 00:02:43.200 align:start position:0% The model displays a 00:02:43.200 --> 00:02:45.710 align:start position:0% The model displays a comparison <00:02:43.491><c>between </c><00:02:43.782><c>actual </c><00:02:44.073><c>and </c><00:02:44.364><c>predicted </c><00:02:44.655><c>inventory </c><00:02:44.946><c>levels, </c><00:02:45.237><c>and</c> 00:02:45.710 --> 00:02:45.720 align:start position:0% comparison between actual and predicted inventory levels, and 00:02:45.720 --> 00:02:48.589 align:start position:0% comparison between actual and predicted inventory levels, and prediction 00:02:48.589 --> 00:02:48.599 align:start position:0% prediction 00:02:48.599 --> 00:02:51.630 align:start position:0% prediction accuracy <00:02:48.752><c>needs </c><00:02:48.905><c>to </c><00:02:49.058><c>be </c><00:02:49.211><c>improved </c><00:02:49.364><c>during </c><00:02:49.517><c>periods </c><00:02:49.670><c>of </c><00:02:49.823><c>rapid </c><00:02:49.976><c>fluctuation. </c><00:02:50.129><c>It </c><00:02:50.282><c>is </c><00:02:50.435><c>important </c><00:02:50.588><c>to </c><00:02:50.741><c>increase </c><00:02:50.894><c>prediction </c><00:02:51.047><c>accuracy </c><00:02:51.200><c>regarding</c> 00:02:51.630 --> 00:02:51.640 align:start position:0% accuracy needs to be improved during periods of rapid fluctuation. It is important to increase prediction accuracy regarding 00:02:51.640 --> 00:02:54.509 align:start position:0% accuracy needs to be improved during periods of rapid fluctuation. It is important to increase prediction accuracy regarding rapid <00:02:51.902><c>fluctuations </c><00:02:52.164><c>through </c><00:02:52.426><c>future </c><00:02:52.688><c>data </c><00:02:52.950><c>acquisition </c><00:02:53.212><c>and </c><00:02:53.474><c>model </c><00:02:53.736><c>improvements. </c><00:02:53.998><c>The</c> 00:02:54.509 --> 00:02:56.550 align:start position:0% rapid fluctuations through future data acquisition and model improvements. The 00:02:56.550 --> 00:03:00.670 align:start position:0% 00:03:00.670 --> 00:03:00.680 align:start position:0% 00:03:00.680 --> 00:03:02.990 align:start position:0% wide <00:03:00.804><c>value </c><00:03:00.928><c>distribution </c><00:03:01.052><c>of </c><00:03:01.176><c>the </c><00:03:01.300><c>month </c><00:03:01.424><c>indicates </c><00:03:01.548><c>that </c><00:03:01.672><c>it </c><00:03:01.796><c>is </c><00:03:01.920><c>an </c><00:03:02.044><c>important </c><00:03:02.168><c>variable. </c><00:03:02.292><c>The </c><00:03:02.416><c>closer </c><00:03:02.540><c>it </c><00:03:02.664><c>is </c><00:03:02.788><c>to</c> 00:03:02.990 --> 00:03:03.000 align:start position:0% wide value distribution of the month indicates that it is an important variable. The closer it is to 00:03:03.000 --> 00:03:05.309 align:start position:0% wide value distribution of the month indicates that it is an important variable. The closer it is to the <00:03:03.336><c>weekend... </c><00:03:03.672><c>Although </c><00:03:04.008><c>the </c><00:03:04.344><c>predicted </c><00:03:04.680><c>value</c> 00:03:05.309 --> 00:03:05.319 align:start position:0% the weekend... Although the predicted value 00:03:05.319 --> 00:03:07.789 align:start position:0% the weekend... Although the predicted value increases, <00:03:05.594><c>the </c><00:03:05.869><c>impact </c><00:03:06.144><c>of </c><00:03:06.419><c>the </c><00:03:06.694><c>number </c><00:03:06.969><c>of </c><00:03:07.244><c>weekdays </c><00:03:07.519><c>is</c> 00:03:07.789 --> 00:03:07.799 align:start position:0% increases, the impact of the number of weekdays is 00:03:07.799 --> 00:03:10.670 align:start position:0% increases, the impact of the number of weekdays is not <00:03:08.067><c>significant </c><00:03:08.335><c>or </c><00:03:08.603><c>acts </c><00:03:08.871><c>in </c><00:03:09.139><c>a </c><00:03:09.407><c>decreasing </c><00:03:09.675><c>direction.</c> 00:03:10.670 --> 00:03:10.680 align:start position:0% not significant or acts in a decreasing direction. 00:03:10.680 --> 00:03:13.550 align:start position:0% not significant or acts in a decreasing direction. Variables <00:03:10.899><c>related </c><00:03:11.118><c>to </c><00:03:11.337><c>anniversaries </c><00:03:11.556><c>do </c><00:03:11.775><c>not </c><00:03:11.994><c>have </c><00:03:12.213><c>a </c><00:03:12.432><c>major </c><00:03:12.651><c>impact </c><00:03:12.870><c>on </c><00:03:13.089><c>the </c><00:03:13.308><c>prediction.</c> 00:03:13.550 --> 00:03:13.560 align:start position:0% Variables related to anniversaries do not have a major impact on the prediction. 00:03:13.560 --> 00:03:16.270 align:start position:0% Variables related to anniversaries do not have a major impact on the prediction. Overall, <00:03:14.840><c>the </c><00:03:16.120><c>two</c> 00:03:16.270 --> 00:03:16.280 align:start position:0% Overall, the two 00:03:16.280 --> 00:03:18.350 align:start position:0% Overall, the two lines <00:03:16.550><c>flow </c><00:03:16.820><c>similarly, </c><00:03:17.090><c>and </c><00:03:17.360><c>the</c> 00:03:18.350 --> 00:03:18.360 align:start position:0% lines flow similarly, and the 00:03:18.360 --> 00:03:21.470 align:start position:0% lines flow similarly, and the stability <00:03:18.800><c>of </c><00:03:19.240><c>the </c><00:03:19.680><c>graph </c><00:03:20.120><c>is </c><00:03:20.560><c>good. </c><00:03:21.000><c>However,</c> 00:03:21.470 --> 00:03:21.480 align:start position:0% stability of the graph is good. However, 00:03:21.480 --> 00:03:23.670 align:start position:0% stability of the graph is good. However, in <00:03:21.930><c>some </c><00:03:22.380><c>sections, </c><00:03:22.830><c>the </c><00:03:23.280><c>prediction</c> 00:03:23.670 --> 00:03:23.680 align:start position:0% in some sections, the prediction 00:03:23.680 --> 00:03:26.550 align:start position:0% in some sections, the prediction appears <00:03:23.894><c>to </c><00:03:24.108><c>be </c><00:03:24.322><c>over- </c><00:03:24.536><c>or </c><00:03:24.750><c>under-predicted. </c><00:03:24.964><c>This </c><00:03:25.178><c>may </c><00:03:25.392><c>be </c><00:03:25.606><c>because </c><00:03:25.820><c>the </c><00:03:26.034><c>model</c> 00:03:26.550 --> 00:03:26.560 align:start position:0% appears to be over- or under-predicted. This may be because the model 00:03:26.560 --> 00:03:29.030 align:start position:0% appears to be over- or under-predicted. This may be because the model reacts <00:03:26.960><c>less </c><00:03:27.360><c>sensitively </c><00:03:27.760><c>to </c><00:03:28.160><c>extreme </c><00:03:28.560><c>values.</c> 00:03:29.030 --> 00:03:32.190 align:start position:0% reacts less sensitively to extreme values. 00:03:32.190 --> 00:03:32.200 align:start position:0% 00:03:32.200 --> 00:03:35.509 align:start position:0% While <00:03:32.531><c>actual </c><00:03:32.862><c>sales </c><00:03:33.193><c>fluctuate </c><00:03:33.524><c>irregularly, </c><00:03:33.855><c>the </c><00:03:34.186><c>predicted </c><00:03:34.517><c>value</c> 00:03:35.509 --> 00:03:35.519 align:start position:0% While actual sales fluctuate irregularly, the predicted value 00:03:35.519 --> 00:03:37.509 align:start position:0% While actual sales fluctuate irregularly, the predicted value remains <00:03:35.799><c>at </c><00:03:36.079><c>an </c><00:03:36.359><c>almost </c><00:03:36.639><c>constant </c><00:03:36.919><c>level.</c> 00:03:37.509 --> 00:03:37.519 align:start position:0% remains at an almost constant level. 00:03:37.519 --> 00:03:40.949 align:start position:0% remains at an almost constant level. In <00:03:37.749><c>other </c><00:03:37.979><c>words, </c><00:03:38.209><c>it </c><00:03:38.439><c>appears </c><00:03:38.669><c>that </c><00:03:38.899><c>the </c><00:03:39.129><c>model </c><00:03:39.359><c>is </c><00:03:39.589><c>only </c><00:03:39.819><c>predicting </c><00:03:40.049><c>the </c><00:03:40.279><c>average </c><00:03:40.509><c>or</c> 00:03:40.949 --> 00:03:40.959 align:start position:0% In other words, it appears that the model is only predicting the average or 00:03:40.959 --> 00:03:43.350 align:start position:0% In other words, it appears that the model is only predicting the average or simple <00:03:42.640><c>value.</c> 00:03:43.350 --> 00:03:43.360 align:start position:0% simple value. 00:03:43.360 --> 00:03:45.470 align:start position:0% simple value. Looking <00:03:43.800><c>at </c><00:03:44.240><c>this </c><00:03:44.680><c>graph, </c><00:03:45.120><c>the</c> 00:03:45.470 --> 00:03:45.480 align:start position:0% Looking at this graph, the 00:03:45.480 --> 00:03:48.229 align:start position:0% Looking at this graph, the predicted <00:03:45.873><c>value </c><00:03:46.266><c>rises </c><00:03:46.659><c>in </c><00:03:47.052><c>early </c><00:03:47.445><c>January </c><00:03:47.838><c>but</c> 00:03:48.229 --> 00:03:48.239 align:start position:0% predicted value rises in early January but 00:03:48.239 --> 00:03:51.110 align:start position:0% predicted value rises in early January but increases <00:03:48.507><c>gradually </c><00:03:48.775><c>from </c><00:03:49.043><c>the </c><00:03:49.311><c>middle </c><00:03:49.579><c>onwards. </c><00:03:49.847><c>However, </c><00:03:50.115><c>towards</c> 00:03:51.110 --> 00:03:51.120 align:start position:0% increases gradually from the middle onwards. However, towards 00:03:51.120 --> 00:03:53.270 align:start position:0% increases gradually from the middle onwards. However, towards the <00:03:51.325><c>end, </c><00:03:51.530><c>it </c><00:03:51.735><c>remains </c><00:03:51.940><c>almost </c><00:03:52.145><c>constant, </c><00:03:52.350><c>showing </c><00:03:52.555><c>a</c> 00:03:53.270 --> 00:03:53.280 align:start position:0% the end, it remains almost constant, showing a 00:03:53.280 --> 00:03:56.149 align:start position:0% the end, it remains almost constant, showing a trend <00:03:53.906><c>of </c><00:03:54.532><c>stabilization. </c><00:03:55.158><c>This</c> 00:03:56.149 --> 00:03:56.159 align:start position:0% trend of stabilization. This 00:03:56.159 --> 00:03:59.030 align:start position:0% trend of stabilization. This graph <00:03:56.411><c>demonstrates </c><00:03:56.663><c>that </c><00:03:56.915><c>the </c><00:03:57.167><c>prediction </c><00:03:57.419><c>results </c><00:03:57.671><c>of </c><00:03:57.923><c>the </c><00:03:58.175><c>Random </c><00:03:58.427><c>Forest </c><00:03:58.679><c>model</c> 00:03:59.030 --> 00:04:01.990 align:start position:0% graph demonstrates that the prediction results of the Random Forest model 00:04:01.990 --> 00:04:02.000 align:start position:0% 00:04:02.000 --> 00:04:04.670 align:start position:0% do <00:04:02.152><c>not </c><00:04:02.304><c>properly </c><00:04:02.456><c>reflect </c><00:04:02.608><c>the </c><00:04:02.760><c>rapid </c><00:04:02.912><c>volatility </c><00:04:03.064><c>of </c><00:04:03.216><c>actual </c><00:04:03.368><c>sales </c><00:04:03.520><c>volume. </c><00:04:03.672><c>The</c> 00:04:04.670 --> 00:04:04.680 align:start position:0% do not properly reflect the rapid volatility of actual sales volume. The 00:04:04.680 --> 00:04:06.750 align:start position:0% do not properly reflect the rapid volatility of actual sales volume. The predicted <00:04:04.829><c>value </c><00:04:04.978><c>appears </c><00:04:05.127><c>as </c><00:04:05.276><c>a </c><00:04:05.425><c>smooth </c><00:04:05.574><c>curve </c><00:04:05.723><c>close </c><00:04:05.872><c>to </c><00:04:06.021><c>the </c><00:04:06.170><c>average </c><00:04:06.319><c>and</c> 00:04:06.750 --> 00:04:09.190 align:start position:0% predicted value appears as a smooth curve close to the average and 00:04:09.190 --> 00:04:09.200 align:start position:0% 00:04:09.200 --> 00:04:12.030 align:start position:0% fails <00:04:09.430><c>to </c><00:04:09.660><c>properly </c><00:04:09.890><c>track </c><00:04:10.120><c>trends </c><00:04:10.350><c>or </c><00:04:10.580><c>sharp </c><00:04:10.810><c>drops. </c><00:04:11.040><c>This</c> 00:04:12.030 --> 00:04:14.509 align:start position:0% fails to properly track trends or sharp drops. This 00:04:14.509 --> 00:04:16.949 align:start position:0% 00:04:16.949 --> 00:04:16.959 align:start position:0% 00:04:16.959 --> 00:04:19.830 align:start position:0% suggests <00:04:17.068><c>that </c><00:04:17.177><c>Random </c><00:04:17.286><c>Forest </c><00:04:17.395><c>exhibits </c><00:04:17.504><c>limitations </c><00:04:17.613><c>in </c><00:04:17.722><c>predicting </c><00:04:17.831><c>irregular </c><00:04:17.940><c>patterns </c><00:04:18.049><c>or </c><00:04:18.158><c>seismic </c><00:04:18.267><c>patterns. </c><00:04:18.376><c>This </c><00:04:18.485><c>graph </c><00:04:18.594><c>shows </c><00:04:18.703><c>that </c><00:04:18.812><c>the </c><00:04:18.921><c>non-predicted </c><00:04:19.030><c>value</c> 00:04:19.830 --> 00:04:22.150 align:start position:0% suggests that Random Forest exhibits limitations in predicting irregular patterns or seismic patterns. This graph shows that the non-predicted value 00:04:22.150 --> 00:04:22.160 align:start position:0% 00:04:22.160 --> 00:04:24.550 align:start position:0% does <00:04:22.270><c>not </c><00:04:22.380><c>properly </c><00:04:22.490><c>reflect </c><00:04:22.600><c>the </c><00:04:22.710><c>large </c><00:04:22.820><c>fluctuations </c><00:04:22.930><c>of </c><00:04:23.040><c>the </c><00:04:23.150><c>actual </c><00:04:23.260><c>value </c><00:04:23.370><c>and </c><00:04:23.480><c>remains </c><00:04:23.590><c>at </c><00:04:23.700><c>the </c><00:04:23.810><c>average </c><00:04:23.920><c>level. </c><00:04:24.030><c>While</c> 00:04:24.550 --> 00:04:24.560 align:start position:0% does not properly reflect the large fluctuations of the actual value and remains at the average level. While 00:04:24.560 --> 00:04:27.070 align:start position:0% does not properly reflect the large fluctuations of the actual value and remains at the average level. While actual 00:04:27.070 --> 00:04:27.080 align:start position:0% actual 00:04:27.080 --> 00:04:29.710 align:start position:0% actual sales <00:04:27.324><c>volume </c><00:04:27.568><c>is </c><00:04:27.812><c>widely </c><00:04:28.056><c>distributed </c><00:04:28.300><c>from </c><00:04:28.544><c>30 </c><00:04:28.788><c>to </c><00:04:29.032><c>180, </c><00:04:29.276><c>the</c> 00:04:29.710 --> 00:04:29.720 align:start position:0% sales volume is widely distributed from 30 to 180, the 00:04:29.720 --> 00:04:32.950 align:start position:0% sales volume is widely distributed from 30 to 180, the predicted <00:04:30.050><c>value </c><00:04:30.380><c>is </c><00:04:30.710><c>in </c><00:04:31.040><c>a </c><00:04:31.370><c>narrow </c><00:04:31.700><c>range </c><00:04:32.030><c>It </c><00:04:32.360><c>is</c> 00:04:32.950 --> 00:04:32.960 align:start position:0% predicted value is in a narrow range It is 00:04:32.960 --> 00:04:35.670 align:start position:0% predicted value is in a narrow range It is concentrated, <00:04:33.346><c>which </c><00:04:33.732><c>means </c><00:04:34.118><c>the </c><00:04:34.504><c>model </c><00:04:34.890><c>has </c><00:04:35.276><c>not</c> 00:04:35.670 --> 00:04:35.680 align:start position:0% concentrated, which means the model has not 00:04:35.680 --> 00:04:37.990 align:start position:0% concentrated, which means the model has not sufficiently <00:04:35.897><c>learned </c><00:04:36.114><c>volatility </c><00:04:36.331><c>such </c><00:04:36.548><c>as </c><00:04:36.765><c>sharp </c><00:04:36.982><c>rises </c><00:04:37.199><c>or </c><00:04:37.416><c>falls. </c><00:04:37.633><c>This</c> 00:04:37.990 --> 00:04:40.510 align:start position:0% sufficiently learned volatility such as sharp rises or falls. This 00:04:40.510 --> 00:04:40.520 align:start position:0% 00:04:40.520 --> 00:04:43.270 align:start position:0% graph <00:04:40.759><c>shows </c><00:04:40.998><c>the </c><00:04:41.237><c>feature </c><00:04:41.476><c>importance </c><00:04:41.715><c>of </c><00:04:41.954><c>the </c><00:04:42.193><c>Random </c><00:04:42.432><c>Forest </c><00:04:42.671><c>model. </c><00:04:42.910><c>The</c> 00:04:43.270 --> 00:04:43.280 align:start position:0% graph shows the feature importance of the Random Forest model. The 00:04:43.280 --> 00:04:45.870 align:start position:0% graph shows the feature importance of the Random Forest model. The most <00:04:45.400><c>important</c> 00:04:45.870 --> 00:04:45.880 align:start position:0% most important 00:04:45.880 --> 00:04:49.070 align:start position:0% most important variable <00:04:46.293><c>is </c><00:04:46.706><c>inventory </c><00:04:47.119><c>level; </c><00:04:47.532><c>its </c><00:04:47.945><c>importance </c><00:04:48.358><c>is</c> 00:04:49.070 --> 00:04:49.080 align:start position:0% variable is inventory level; its importance is 00:04:49.080 --> 00:04:51.710 align:start position:0% variable is inventory level; its importance is overwhelmingly <00:04:49.600><c>high </c><00:04:50.120><c>at </c><00:04:50.640><c>over </c><00:04:51.160><c>0.7,</c> 00:04:51.710 --> 00:04:51.720 align:start position:0% overwhelmingly high at over 0.7, 00:04:51.720 --> 00:04:55.390 align:start position:0% overwhelmingly high at over 0.7, having <00:04:52.090><c>the </c><00:04:52.460><c>greatest </c><00:04:52.830><c>impact </c><00:04:53.200><c>on </c><00:04:53.570><c>sales </c><00:04:53.940><c>volume </c><00:04:54.310><c>prediction. </c><00:04:54.680><c>Weather,</c> 00:04:55.390 --> 00:04:55.400 align:start position:0% having the greatest impact on sales volume prediction. Weather, 00:04:55.400 --> 00:04:58.550 align:start position:0% having the greatest impact on sales volume prediction. Weather, price, <00:04:55.806><c>event </c><00:04:56.212><c>status, </c><00:04:56.618><c>and </c><00:04:57.024><c>competitor </c><00:04:57.430><c>prices </c><00:04:57.836><c>also</c> 00:04:58.550 --> 00:05:00.189 align:start position:0% price, event status, and competitor prices also 00:05:00.189 --> 00:05:00.199 align:start position:0% 00:05:00.199 --> 00:05:03.629 align:start position:0% appeared <00:05:00.346><c>as </c><00:05:00.493><c>variables </c><00:05:00.640><c>affecting </c><00:05:00.787><c>demand, </c><00:05:00.934><c>but </c><00:05:01.081><c>they </c><00:05:01.228><c>are </c><00:05:01.375><c>likely </c><00:05:01.522><c>to </c><00:05:01.669><c>have </c><00:05:01.816><c>a </c><00:05:01.963><c>greater </c><00:05:02.110><c>impact </c><00:05:02.257><c>when </c><00:05:02.404><c>combined </c><00:05:02.551><c>rather </c><00:05:02.698><c>than </c><00:05:02.845><c>individually. </c><00:05:02.992><c>This</c> 00:05:03.629 --> 00:05:06.590 align:start position:0% appeared as variables affecting demand, but they are likely to have a greater impact when combined rather than individually. This 00:05:06.590 --> 00:05:06.600 align:start position:0% 00:05:06.600 --> 00:05:09.629 align:start position:0% graph <00:05:06.793><c>visually </c><00:05:06.986><c>displays </c><00:05:07.179><c>not </c><00:05:07.372><c>only </c><00:05:07.565><c>the </c><00:05:07.758><c>influence </c><00:05:07.951><c>of </c><00:05:08.144><c>each </c><00:05:08.337><c>variable </c><00:05:08.530><c>but </c><00:05:08.723><c>also </c><00:05:08.916><c>the</c> 00:05:09.629 --> 00:05:09.639 align:start position:0% graph visually displays not only the influence of each variable but also the 00:05:09.639 --> 00:05:12.710 align:start position:0% graph visually displays not only the influence of each variable but also the direction <00:05:09.883><c>and </c><00:05:10.127><c>magnitude </c><00:05:10.371><c>of </c><00:05:10.615><c>their </c><00:05:10.859><c>impact </c><00:05:11.103><c>on </c><00:05:11.347><c>the </c><00:05:11.591><c>predicted </c><00:05:11.835><c>values.</c> 00:05:12.710 --> 00:05:12.720 align:start position:0% direction and magnitude of their impact on the predicted values. 00:05:12.720 --> 00:05:16.189 align:start position:0% direction and magnitude of their impact on the predicted values. Inventory <00:05:13.600><c>level, </c><00:05:14.480><c>price, </c><00:05:15.360><c>and</c> 00:05:16.189 --> 00:05:16.199 align:start position:0% Inventory level, price, and 00:05:16.199 --> 00:05:18.309 align:start position:0% Inventory level, price, and event <00:05:16.421><c>status </c><00:05:16.643><c>emerged </c><00:05:16.865><c>as </c><00:05:17.087><c>the </c><00:05:17.309><c>most </c><00:05:17.531><c>important </c><00:05:17.753><c>variables</c> 00:05:18.309 --> 00:05:18.319 align:start position:0% event status emerged as the most important variables 00:05:18.319 --> 00:05:21.029 align:start position:0% event status emerged as the most important variables ; 00:05:21.029 --> 00:05:21.039 align:start position:0% ; 00:05:21.039 --> 00:05:24.070 align:start position:0% ; predicted <00:05:21.336><c>values </c><00:05:21.633><c>tended </c><00:05:21.930><c>to </c><00:05:22.227><c>decrease </c><00:05:22.524><c>as </c><00:05:22.821><c>inventory </c><00:05:23.118><c>levels </c><00:05:23.415><c>increased </c><00:05:23.712><c>and</c> 00:05:24.070 --> 00:05:24.080 align:start position:0% predicted values tended to decrease as inventory levels increased and 00:05:24.080 --> 00:05:26.710 align:start position:0% predicted values tended to decrease as inventory levels increased and increase <00:05:24.314><c>as </c><00:05:24.548><c>the </c><00:05:24.782><c>number </c><00:05:25.016><c>of </c><00:05:25.250><c>events </c><00:05:25.484><c>increased. </c><00:05:25.718><c>This</c> 00:05:26.710 --> 00:05:29.670 align:start position:0% increase as the number of events increased. This 00:05:29.670 --> 00:05:29.680 align:start position:0% 00:05:29.680 --> 00:05:32.150 align:start position:0% confirms <00:05:29.843><c>that </c><00:05:30.006><c>the </c><00:05:30.169><c>model </c><00:05:30.332><c>is </c><00:05:30.495><c>effectively </c><00:05:30.658><c>learning </c><00:05:30.821><c>the </c><00:05:30.984><c>factors </c><00:05:31.147><c>that </c><00:05:31.310><c>influence </c><00:05:31.473><c>actual </c><00:05:31.636><c>demand. </c><00:05:31.799><c>This</c> 00:05:32.150 --> 00:05:32.160 align:start position:0% confirms that the model is effectively learning the factors that influence actual demand. This 00:05:32.160 --> 00:05:35.070 align:start position:0% confirms that the model is effectively learning the factors that influence actual demand. This graph <00:05:32.809><c>shows </c><00:05:33.458><c>the </c><00:05:34.107><c>changes </c><00:05:34.756><c>in</c> 00:05:35.070 --> 00:05:35.080 align:start position:0% graph shows the changes in 00:05:35.080 --> 00:05:37.670 align:start position:0% graph shows the changes in training <00:05:35.296><c>loss </c><00:05:35.512><c>and </c><00:05:35.728><c>validation </c><00:05:35.944><c>loss </c><00:05:36.160><c>during </c><00:05:36.376><c>the </c><00:05:36.592><c>RF </c><00:05:36.808><c>model </c><00:05:37.024><c>training </c><00:05:37.240><c>process.</c> 00:05:37.670 --> 00:05:40.309 align:start position:0% training loss and validation loss during the RF model training process. 00:05:40.309 --> 00:05:40.319 align:start position:0% 00:05:40.319 --> 00:05:43.510 align:start position:0% Performance <00:05:40.537><c>improved </c><00:05:40.755><c>rapidly </c><00:05:40.973><c>as </c><00:05:41.191><c>the </c><00:05:41.409><c>number </c><00:05:41.627><c>of </c><00:05:41.845><c>trees </c><00:05:42.063><c>increased, </c><00:05:42.281><c>and </c><00:05:42.499><c>it </c><00:05:42.717><c>showed </c><00:05:42.935><c>stable </c><00:05:43.153><c>convergence</c> 00:05:43.510 --> 00:05:43.520 align:start position:0% Performance improved rapidly as the number of trees increased, and it showed stable convergence 00:05:43.520 --> 00:05:45.950 align:start position:0% Performance improved rapidly as the number of trees increased, and it showed stable convergence after <00:05:43.715><c>approximately </c><00:05:43.910><c>20 </c><00:05:44.105><c>to </c><00:05:44.300><c>30 </c><00:05:44.495><c>trees. </c><00:05:44.690><c>Since </c><00:05:44.885><c>there </c><00:05:45.080><c>was</c> 00:05:45.950 --> 00:05:48.670 align:start position:0% after approximately 20 to 30 trees. Since there was 00:05:48.670 --> 00:05:48.680 align:start position:0% 00:05:48.680 --> 00:05:51.670 align:start position:0% almost <00:05:48.977><c>no </c><00:05:49.274><c>difference </c><00:05:49.571><c>between </c><00:05:49.868><c>training </c><00:05:50.165><c>loss </c><00:05:50.462><c>and </c><00:05:50.759><c>validation </c><00:05:51.056><c>loss, </c><00:05:51.353><c>it</c> 00:05:51.670 --> 00:05:53.909 align:start position:0% almost no difference between training loss and validation loss, it 00:05:53.909 --> 00:05:53.919 align:start position:0% 00:05:53.919 --> 00:05:56.670 align:start position:0% can <00:05:54.095><c>be </c><00:05:54.271><c>confirmed </c><00:05:54.447><c>that </c><00:05:54.623><c>the </c><00:05:54.799><c>model </c><00:05:54.975><c>trained </c><00:05:55.151><c>stably </c><00:05:55.327><c>without </c><00:05:55.503><c>overfitting. </c><00:05:55.679><c>This</c> 00:05:56.670 --> 00:05:56.680 align:start position:0% can be confirmed that the model trained stably without overfitting. This 00:05:56.680 --> 00:05:58.710 align:start position:0% can be confirmed that the model trained stably without overfitting. This graph <00:05:57.380><c>shows </c><00:05:58.080><c>the</c> 00:05:58.710 --> 00:05:58.720 align:start position:0% graph shows the 00:05:58.720 --> 00:06:00.830 align:start position:0% graph shows the results <00:05:58.858><c>of </c><00:05:58.996><c>predicting </c><00:05:59.134><c>10 </c><00:05:59.272><c>days </c><00:05:59.410><c>of </c><00:05:59.548><c>data </c><00:05:59.686><c>using </c><00:05:59.824><c>the </c><00:05:59.962><c>RF </c><00:06:00.100><c>model. </c><00:06:00.238><c>The</c> 00:06:00.830 --> 00:06:00.840 align:start position:0% results of predicting 10 days of data using the RF model. The 00:06:00.840 --> 00:06:03.749 align:start position:0% results of predicting 10 days of data using the RF model. The actual <00:06:01.479><c>and </c><00:06:02.118><c>predicted </c><00:06:02.757><c>values</c> 00:06:03.749 --> 00:06:03.759 align:start position:0% actual and predicted values 00:06:03.759 --> 00:06:05.830 align:start position:0% actual and predicted values generally <00:06:04.139><c>show </c><00:06:04.519><c>similar </c><00:06:04.899><c>patterns, </c><00:06:05.279><c>with</c> 00:06:05.830 --> 00:06:05.840 align:start position:0% generally show similar patterns, with 00:06:05.840 --> 00:06:08.309 align:start position:0% generally show similar patterns, with upward <00:06:06.091><c>and </c><00:06:06.342><c>downward </c><00:06:06.593><c>flows </c><00:06:06.844><c>being </c><00:06:07.095><c>almost </c><00:06:07.346><c>identical. </c><00:06:07.597><c>It</c> 00:06:08.309 --> 00:06:08.319 align:start position:0% upward and downward flows being almost identical. It 00:06:08.319 --> 00:06:11.749 align:start position:0% upward and downward flows being almost identical. It follows <00:06:09.699><c>the </c><00:06:11.079><c>trend</c> 00:06:11.749 --> 00:06:11.759 align:start position:0% follows the trend 00:06:11.759 --> 00:06:13.830 align:start position:0% follows the trend well, <00:06:11.884><c>particularly </c><00:06:12.009><c>reflecting </c><00:06:12.134><c>periods </c><00:06:12.259><c>of </c><00:06:12.384><c>rapid </c><00:06:12.509><c>change, </c><00:06:12.634><c>confirming </c><00:06:12.759><c>that </c><00:06:12.884><c>the </c><00:06:13.009><c>model </c><00:06:13.134><c>has </c><00:06:13.259><c>effectively </c><00:06:13.384><c>learned </c><00:06:13.509><c>short-term</c> 00:06:13.830 --> 00:06:13.840 align:start position:0% well, particularly reflecting periods of rapid change, confirming that the model has effectively learned short-term 00:06:13.840 --> 00:06:16.629 align:start position:0% well, particularly reflecting periods of rapid change, confirming that the model has effectively learned short-term trends <00:06:14.410><c>and </c><00:06:14.980><c>pattern </c><00:06:15.550><c>sensitivity. </c><00:06:16.120><c>This</c> 00:06:16.629 --> 00:06:16.639 align:start position:0% trends and pattern sensitivity. This 00:06:16.639 --> 00:06:19.390 align:start position:0% trends and pattern sensitivity. This graph 00:06:19.390 --> 00:06:21.230 align:start position:0% graph 00:06:21.230 --> 00:06:23.749 align:start position:0% 00:06:23.749 --> 00:06:23.759 align:start position:0% 00:06:23.759 --> 00:06:26.950 align:start position:0% shows <00:06:23.875><c>the </c><00:06:23.991><c>results </c><00:06:24.107><c>of </c><00:06:24.223><c>comparing </c><00:06:24.339><c>actual </c><00:06:24.455><c>and </c><00:06:24.571><c>predicted </c><00:06:24.687><c>values </c><00:06:24.803><c>over </c><00:06:24.919><c>10 </c><00:06:25.035><c>days </c><00:06:25.151><c>for </c><00:06:25.267><c>five </c><00:06:25.383><c>samples. </c><00:06:25.499><c>In </c><00:06:25.615><c>all </c><00:06:25.731><c>samples, </c><00:06:25.847><c>the </c><00:06:25.963><c>blue </c><00:06:26.079><c>and</c> 00:06:26.950 --> 00:06:26.960 align:start position:0% shows the results of comparing actual and predicted values over 10 days for five samples. In all samples, the blue and 00:06:26.960 --> 00:06:29.950 align:start position:0% shows the results of comparing actual and predicted values over 10 days for five samples. In all samples, the blue and orange <00:06:27.533><c>lines </c><00:06:28.106><c>almost </c><00:06:28.679><c>overlap,</c> 00:06:29.950 --> 00:06:29.960 align:start position:0% orange lines almost overlap, 00:06:29.960 --> 00:06:32.390 align:start position:0% orange lines almost overlap, confirming <00:06:30.182><c>that </c><00:06:30.404><c>the </c><00:06:30.626><c>predicted </c><00:06:30.848><c>values </c><00:06:31.070><c>closely </c><00:06:31.292><c>follow </c><00:06:31.514><c>the </c><00:06:31.736><c>actual </c><00:06:31.958><c>values.</c> 00:06:32.390 --> 00:06:32.400 align:start position:0% confirming that the predicted values closely follow the actual values. 00:06:32.400 --> 00:06:35.309 align:start position:0% confirming that the predicted values closely follow the actual values. In <00:06:32.670><c>particular, </c><00:06:32.940><c>not </c><00:06:33.210><c>only </c><00:06:33.480><c>the </c><00:06:33.750><c>trend </c><00:06:34.020><c>but </c><00:06:34.290><c>also </c><00:06:34.560><c>the</c> 00:06:35.309 --> 00:06:35.319 align:start position:0% In particular, not only the trend but also the 00:06:35.319 --> 00:06:38.029 align:start position:0% In particular, not only the trend but also the locations <00:06:35.564><c>of </c><00:06:35.809><c>the </c><00:06:36.054><c>highs </c><00:06:36.299><c>and </c><00:06:36.544><c>lows </c><00:06:36.789><c>are </c><00:06:37.034><c>nearly </c><00:06:37.279><c>identical,</c> 00:06:38.029 --> 00:06:38.039 align:start position:0% locations of the highs and lows are nearly identical, 00:06:38.039 --> 00:06:40.710 align:start position:0% locations of the highs and lows are nearly identical, indicating <00:06:38.349><c>the </c><00:06:38.659><c>model's </c><00:06:38.969><c>excellent </c><00:06:39.279><c>pattern </c><00:06:39.589><c>learning </c><00:06:39.899><c>performance. </c><00:06:40.209><c>Although </c><00:06:40.519><c>the</c> 00:06:40.710 --> 00:06:40.720 align:start position:0% indicating the model's excellent pattern learning performance. Although the 00:06:40.720 --> 00:06:43.309 align:start position:0% indicating the model's excellent pattern learning performance. Although the Random <00:06:41.800><c>Forest </c><00:06:42.880><c>model</c> 00:06:43.309 --> 00:06:43.319 align:start position:0% Random Forest model 00:06:43.319 --> 00:06:46.309 align:start position:0% Random Forest model shows <00:06:43.569><c>a </c><00:06:43.819><c>flow </c><00:06:44.069><c>similar </c><00:06:44.319><c>to </c><00:06:44.569><c>the </c><00:06:44.819><c>actual </c><00:06:45.069><c>values, </c><00:06:45.319><c>the</c> 00:06:46.309 --> 00:06:46.319 align:start position:0% shows a flow similar to the actual values, the 00:06:46.319 --> 00:06:49.309 align:start position:0% shows a flow similar to the actual values, the fluctuation <00:06:46.775><c>range </c><00:06:47.231><c>of </c><00:06:47.687><c>the </c><00:06:48.143><c>predicted </c><00:06:48.599><c>values</c> 00:06:49.309 --> 00:06:49.319 align:start position:0% fluctuation range of the predicted values 00:06:49.319 --> 00:06:51.670 align:start position:0% fluctuation range of the predicted values appears <00:06:49.492><c>larger </c><00:06:49.665><c>than </c><00:06:49.838><c>the </c><00:06:50.011><c>actual </c><00:06:50.184><c>values </c><00:06:50.357><c>in </c><00:06:50.530><c>some </c><00:06:50.703><c>sections. </c><00:06:50.876><c>This</c> 00:06:51.670 --> 00:06:54.110 align:start position:0% appears larger than the actual values in some sections. This 00:06:54.110 --> 00:06:56.670 align:start position:0% 00:06:56.670 --> 00:06:56.680 align:start position:0% 00:06:56.680 --> 00:06:59.550 align:start position:0% phenomenon <00:06:56.835><c>occurs </c><00:06:56.990><c>because </c><00:06:57.145><c>Random </c><00:06:57.300><c>Forest </c><00:06:57.455><c>tends </c><00:06:57.610><c>to </c><00:06:57.765><c>predict </c><00:06:57.920><c>extreme </c><00:06:58.075><c>values </c><00:06:58.230><c>in </c><00:06:58.385><c>certain </c><00:06:58.540><c>feature </c><00:06:58.695><c>combinations. </c><00:06:58.850><c>The </c><00:06:59.005><c>overall </c><00:06:59.160><c>demand</c> 00:06:59.550 --> 00:06:59.560 align:start position:0% phenomenon occurs because Random Forest tends to predict extreme values in certain feature combinations. The overall demand 00:06:59.560 --> 00:07:02.309 align:start position:0% phenomenon occurs because Random Forest tends to predict extreme values in certain feature combinations. The overall demand pattern <00:06:59.862><c>has </c><00:07:00.164><c>been </c><00:07:00.466><c>learned </c><00:07:00.768><c>well. </c><00:07:01.070><c>Some </c><00:07:01.372><c>predicted </c><00:07:01.674><c>values</c> 00:07:02.309 --> 00:07:02.319 align:start position:0% pattern has been learned well. Some predicted values 00:07:02.319 --> 00:07:05.070 align:start position:0% pattern has been learned well. Some predicted values exhibit <00:07:02.539><c>characteristics </c><00:07:02.759><c>of </c><00:07:02.979><c>being </c><00:07:03.199><c>excessively </c><00:07:03.419><c>high </c><00:07:03.639><c>or </c><00:07:03.859><c>low. </c><00:07:04.079><c>These</c> 00:07:05.070 --> 00:07:05.080 align:start position:0% exhibit characteristics of being excessively high or low. These 00:07:05.080 --> 00:07:08.270 align:start position:0% exhibit characteristics of being excessively high or low. These are <00:07:05.355><c>the </c><00:07:05.630><c>prediction </c><00:07:05.905><c>results </c><00:07:06.180><c>for </c><00:07:06.455><c>the </c><00:07:06.730><c>data. </c><00:07:07.005><c>Since </c><00:07:07.280><c>the</c> 00:07:08.270 --> 00:07:08.280 align:start position:0% are the prediction results for the data. Since the 00:07:08.280 --> 00:07:10.749 align:start position:0% are the prediction results for the data. Since the actual <00:07:08.531><c>and </c><00:07:08.782><c>predicted </c><00:07:09.033><c>values </c><00:07:09.284><c>match </c><00:07:09.535><c>almost </c><00:07:09.786><c>perfectly, </c><00:07:10.037><c>it</c> 00:07:10.749 --> 00:07:10.759 align:start position:0% actual and predicted values match almost perfectly, it 00:07:10.759 --> 00:07:13.029 align:start position:0% actual and predicted values match almost perfectly, it accurately <00:07:11.103><c>predicts </c><00:07:11.447><c>upward </c><00:07:11.791><c>and </c><00:07:12.135><c>downward </c><00:07:12.479><c>patterns.</c> 00:07:13.029 --> 00:07:13.039 align:start position:0% accurately predicts upward and downward patterns. 00:07:13.039 --> 00:07:16.510 align:start position:0% accurately predicts upward and downward patterns. In <00:07:14.519><c>particular, </c><00:07:15.999><c>it</c> 00:07:16.510 --> 00:07:16.520 align:start position:0% In particular, it 00:07:16.520 --> 00:07:19.230 align:start position:0% In particular, it demonstrates <00:07:16.710><c>high </c><00:07:16.900><c>accuracy </c><00:07:17.090><c>in </c><00:07:17.280><c>both </c><00:07:17.470><c>rapid </c><00:07:17.660><c>rise </c><00:07:17.850><c>and </c><00:07:18.040><c>stable </c><00:07:18.230><c>maintenance </c><00:07:18.420><c>periods, </c><00:07:18.610><c>confirming </c><00:07:18.800><c>that</c> 00:07:19.230 --> 00:07:19.240 align:start position:0% demonstrates high accuracy in both rapid rise and stable maintenance periods, confirming that 00:07:19.240 --> 00:07:22.390 align:start position:0% demonstrates high accuracy in both rapid rise and stable maintenance periods, confirming that the <00:07:19.668><c>model </c><00:07:20.096><c>has </c><00:07:20.524><c>learned </c><00:07:20.952><c>the </c><00:07:21.380><c>patterns </c><00:07:21.808><c>well. </c><00:07:22.236><c>The</c> 00:07:22.390 --> 00:07:24.909 align:start position:0% the model has learned the patterns well. The 00:07:24.909 --> 00:07:24.919 align:start position:0% 00:07:24.919 --> 00:07:26.990 align:start position:0% sales <00:07:25.033><c>volume </c><00:07:25.147><c>prediction </c><00:07:25.261><c>results </c><00:07:25.375><c>for </c><00:07:25.489><c>10 </c><00:07:25.603><c>days </c><00:07:25.717><c>from </c><00:07:25.831><c>Sample </c><00:07:25.945><c>0 </c><00:07:26.059><c>to </c><00:07:26.173><c>Sample </c><00:07:26.287><c>4 </c><00:07:26.401><c>have </c><00:07:26.515><c>been </c><00:07:26.629><c>visualized</c> 00:07:26.990 --> 00:07:27.000 align:start position:0% sales volume prediction results for 10 days from Sample 0 to Sample 4 have been visualized 00:07:27.000 --> 00:07:28.510 align:start position:0% sales volume prediction results for 10 days from Sample 0 to Sample 4 have been visualized in a single graph. The 00:07:28.510 --> 00:07:28.520 align:start position:0% in a single graph. The 00:07:28.520 --> 00:07:30.510 align:start position:0% in a single graph. The actual <00:07:28.986><c>and </c><00:07:29.452><c>predicted </c><00:07:29.918><c>values</c> 00:07:30.510 --> 00:07:30.520 align:start position:0% actual and predicted values 00:07:30.520 --> 00:07:33.629 align:start position:0% actual and predicted values appear <00:07:30.805><c>mostly </c><00:07:31.090><c>similar, </c><00:07:31.375><c>and </c><00:07:31.660><c>in </c><00:07:31.945><c>particular, </c><00:07:32.230><c>the </c><00:07:32.515><c>upward </c><00:07:32.800><c>and</c> 00:07:33.629 --> 00:07:33.639 align:start position:0% appear mostly similar, and in particular, the upward and 00:07:33.639 --> 00:07:36.430 align:start position:0% appear mostly similar, and in particular, the upward and downward <00:07:34.007><c>patterns </c><00:07:34.375><c>match </c><00:07:34.743><c>almost </c><00:07:35.111><c>perfectly. </c><00:07:35.479><c>This</c> 00:07:36.430 --> 00:07:36.440 align:start position:0% downward patterns match almost perfectly. This 00:07:36.440 --> 00:07:39.550 align:start position:0% downward patterns match almost perfectly. This demonstrates <00:07:36.705><c>the </c><00:07:36.970><c>excellent </c><00:07:37.235><c>short-term </c><00:07:37.500><c>forecasting </c><00:07:37.765><c>performance </c><00:07:38.030><c>of </c><00:07:38.295><c>the </c><00:07:38.560><c>model.</c> 00:07:39.550 --> 00:07:39.560 align:start position:0% demonstrates the excellent short-term forecasting performance of the model. 00:07:39.560 --> 00:07:42.189 align:start position:0% demonstrates the excellent short-term forecasting performance of the model. The <00:07:39.819><c>model </c><00:07:40.078><c>exhibits </c><00:07:40.337><c>consistent </c><00:07:40.596><c>performance </c><00:07:40.855><c>across </c><00:07:41.114><c>various </c><00:07:41.373><c>samples </c><00:07:41.632><c>and</c> 00:07:42.189 --> 00:07:44.230 align:start position:0% The model exhibits consistent performance across various samples and 00:07:44.230 --> 00:07:44.240 align:start position:0% 00:07:44.240 --> 00:07:47.469 align:start position:0% accurately <00:07:44.969><c>reflects </c><00:07:45.698><c>changes </c><00:07:46.427><c>over </c><00:07:47.156><c>time.</c> 00:07:47.469 --> 00:07:47.479 align:start position:0% accurately reflects changes over time. 00:07:47.479 --> 00:07:50.230 align:start position:0% accurately reflects changes over time. Comparing <00:07:47.645><c>the </c><00:07:47.811><c>average </c><00:07:47.977><c>of </c><00:07:48.143><c>the </c><00:07:48.309><c>predicted </c><00:07:48.475><c>and </c><00:07:48.641><c>actual </c><00:07:48.807><c>values </c><00:07:48.973><c>across </c><00:07:49.139><c>the </c><00:07:49.305><c>entire </c><00:07:49.471><c>sample, </c><00:07:49.637><c>the</c> 00:07:50.230 --> 00:07:50.240 align:start position:0% Comparing the average of the predicted and actual values across the entire sample, the 00:07:50.240 --> 00:07:52.909 align:start position:0% Comparing the average of the predicted and actual values across the entire sample, the RF <00:07:50.534><c>model </c><00:07:50.828><c>demonstrated </c><00:07:51.122><c>accurate </c><00:07:51.416><c>and </c><00:07:51.710><c>stable </c><00:07:52.004><c>performance </c><00:07:52.298><c>not </c><00:07:52.592><c>only</c> 00:07:52.909 --> 00:07:52.919 align:start position:0% RF model demonstrated accurate and stable performance not only 00:07:52.919 --> 00:07:55.550 align:start position:0% RF model demonstrated accurate and stable performance not only in <00:07:53.105><c>individual </c><00:07:53.291><c>predictions </c><00:07:53.477><c>but </c><00:07:53.663><c>also </c><00:07:53.849><c>in </c><00:07:54.035><c>the </c><00:07:54.221><c>overall </c><00:07:54.407><c>average. </c><00:07:54.593><c>It</c> 00:07:55.550 --> 00:07:58.510 align:start position:0% in individual predictions but also in the overall average. It 00:07:58.510 --> 00:08:01.110 align:start position:0% 00:08:01.110 --> 00:08:03.749 align:start position:0% 00:08:03.749 --> 00:08:06.189 align:start position:0% 00:08:06.189 --> 00:08:06.199 align:start position:0% 00:08:06.199 --> 00:08:08.189 align:start position:0% confirms <00:08:06.286><c>that </c><00:08:06.373><c>the </c><00:08:06.460><c>model </c><00:08:06.547><c>possesses </c><00:08:06.634><c>high </c><00:08:06.721><c>average </c><00:08:06.808><c>reliability, </c><00:08:06.895><c>as </c><00:08:06.982><c>predictions </c><00:08:07.069><c>maintain </c><00:08:07.156><c>a </c><00:08:07.243><c>consistent </c><00:08:07.330><c>scale </c><00:08:07.417><c>without </c><00:08:07.504><c>consistently </c><00:08:07.591><c>underestimating </c><00:08:07.678><c>or </c><00:08:07.765><c>overestimating </c><00:08:07.852><c>within </c><00:08:07.939><c>specific </c><00:08:08.026><c>ranges.</c> 00:08:08.189 --> 00:08:08.199 align:start position:0% confirms that the model possesses high average reliability, as predictions maintain a consistent scale without consistently underestimating or overestimating within specific ranges. 00:08:08.199 --> 00:08:11.830 align:start position:0% confirms that the model possesses high average reliability, as predictions maintain a consistent scale without consistently underestimating or overestimating within specific ranges. These <00:08:08.665><c>results </c><00:08:09.131><c>indicate </c><00:08:09.597><c>that </c><00:08:10.063><c>the </c><00:08:10.529><c>RF </c><00:08:10.995><c>model</c> 00:08:11.830 --> 00:08:14.230 align:start position:0% These results indicate that the RF model 00:08:14.230 --> 00:08:14.240 align:start position:0% 00:08:14.240 --> 00:08:17.070 align:start position:0% demonstrates <00:08:14.502><c>overall </c><00:08:14.764><c>superior </c><00:08:15.026><c>performance </c><00:08:15.288><c>in </c><00:08:15.550><c>demand </c><00:08:15.812><c>forecasting. </c><00:08:16.074><c>This</c> 00:08:17.070 --> 00:08:17.080 align:start position:0% demonstrates overall superior performance in demand forecasting. This 00:08:17.080 --> 00:08:20.350 align:start position:0% demonstrates overall superior performance in demand forecasting. This graph <00:08:17.583><c>shows </c><00:08:18.086><c>the </c><00:08:18.589><c>results </c><00:08:19.092><c>of </c><00:08:19.595><c>a</c> 00:08:20.350 --> 00:08:20.360 align:start position:0% graph shows the results of a 00:08:20.360 --> 00:08:23.469 align:start position:0% graph shows the results of a scenario <00:08:20.530><c>analysis </c><00:08:20.700><c>of </c><00:08:20.870><c>sales </c><00:08:21.040><c>volume </c><00:08:21.210><c>changes </c><00:08:21.380><c>over </c><00:08:21.550><c>the </c><00:08:21.720><c>next </c><00:08:21.890><c>10 </c><00:08:22.060><c>days, </c><00:08:22.230><c>varying </c><00:08:22.400><c>prices </c><00:08:22.570><c>and </c><00:08:22.740><c>inventory </c><00:08:22.910><c>levels.</c> 00:08:23.469 --> 00:08:26.309 align:start position:0% scenario analysis of sales volume changes over the next 10 days, varying prices and inventory levels. 00:08:26.309 --> 00:08:26.319 align:start position:0% 00:08:26.319 --> 00:08:28.629 align:start position:0% Predicted <00:08:27.239><c>sales </c><00:08:28.159><c>volume</c> 00:08:28.629 --> 00:08:28.639 align:start position:0% Predicted sales volume 00:08:28.639 --> 00:08:30.670 align:start position:0% Predicted sales volume tends <00:08:28.705><c>to </c><00:08:28.771><c>increase </c><00:08:28.837><c>as </c><00:08:28.903><c>prices </c><00:08:28.969><c>drop </c><00:08:29.035><c>and </c><00:08:29.101><c>inventory </c><00:08:29.167><c>levels </c><00:08:29.233><c>rise; </c><00:08:29.299><c>in </c><00:08:29.365><c>particular, </c><00:08:29.431><c>the </c><00:08:29.497><c>highest </c><00:08:29.563><c>sales </c><00:08:29.629><c>volume </c><00:08:29.695><c>is </c><00:08:29.761><c>recorded </c><00:08:29.827><c>when </c><00:08:29.893><c>the </c><00:08:29.959><c>price </c><00:08:30.025><c>is</c> 00:08:30.670 --> 00:08:30.680 align:start position:0% tends to increase as prices drop and inventory levels rise; in particular, the highest sales volume is recorded when the price is 00:08:30.680 --> 00:08:32.790 align:start position:0% tends to increase as prices drop and inventory levels rise; in particular, the highest sales volume is recorded when the price is 15,000 <00:08:30.913><c>won </c><00:08:31.146><c>and </c><00:08:31.379><c>inventory </c><00:08:31.612><c>is </c><00:08:31.845><c>150 </c><00:08:32.078><c>units.</c> 00:08:32.790 --> 00:08:35.190 align:start position:0% 15,000 won and inventory is 150 units. 00:08:35.190 --> 00:08:35.200 align:start position:0% 00:08:35.200 --> 00:08:38.589 align:start position:0% Conversely, 00:08:38.589 --> 00:08:38.599 align:start position:0% Conversely, 00:08:38.599 --> 00:08:41.709 align:start position:0% Conversely, sales <00:08:38.763><c>volume </c><00:08:38.927><c>plummets </c><00:08:39.091><c>when </c><00:08:39.255><c>the </c><00:08:39.419><c>price </c><00:08:39.583><c>is </c><00:08:39.747><c>35,000 </c><00:08:39.911><c>won, </c><00:08:40.075><c>confirming </c><00:08:40.239><c>that </c><00:08:40.403><c>demand </c><00:08:40.567><c>elasticity </c><00:08:40.731><c>plays </c><00:08:40.895><c>a </c><00:08:41.059><c>significant </c><00:08:41.223><c>role. </c><00:08:41.387><c>This</c> 00:08:41.709 --> 00:08:44.230 align:start position:0% sales volume plummets when the price is 35,000 won, confirming that demand elasticity plays a significant role. This 00:08:44.230 --> 00:08:44.240 align:start position:0% 00:08:44.240 --> 00:08:47.190 align:start position:0% graph <00:08:44.626><c>visualizes </c><00:08:45.012><c>the </c><00:08:45.398><c>importance </c><00:08:45.784><c>of </c><00:08:46.170><c>permutation </c><00:08:46.556><c>characteristics.</c> 00:08:47.190 --> 00:08:47.200 align:start position:0% graph visualizes the importance of permutation characteristics. 00:08:47.200 --> 00:08:50.269 align:start position:0% graph visualizes the importance of permutation characteristics. Event <00:08:48.119><c>presence </c><00:08:49.038><c>and </c><00:08:49.957><c>price</c> 00:08:50.269 --> 00:08:50.279 align:start position:0% Event presence and price 00:08:50.279 --> 00:08:52.590 align:start position:0% Event presence and price emerged <00:08:50.536><c>as </c><00:08:50.793><c>the </c><00:08:51.050><c>most </c><00:08:51.307><c>important </c><00:08:51.564><c>characteristics, </c><00:08:51.821><c>while </c><00:08:52.078><c>competitor</c> 00:08:52.590 --> 00:08:52.600 align:start position:0% emerged as the most important characteristics, while competitor 00:08:52.600 --> 00:08:55.389 align:start position:0% emerged as the most important characteristics, while competitor pricing <00:08:52.925><c>and </c><00:08:53.250><c>lead </c><00:08:53.575><c>time </c><00:08:53.900><c>also </c><00:08:54.225><c>showed </c><00:08:54.550><c>relatively </c><00:08:54.875><c>high</c> 00:08:55.389 --> 00:08:55.399 align:start position:0% pricing and lead time also showed relatively high 00:08:55.399 --> 00:08:58.750 align:start position:0% pricing and lead time also showed relatively high contributions. <00:08:55.865><c>On </c><00:08:56.331><c>the </c><00:08:56.797><c>other </c><00:08:57.263><c>hand, </c><00:08:57.729><c>seasonal </c><00:08:58.195><c>variables</c> 00:08:58.750 --> 00:08:58.760 align:start position:0% contributions. On the other hand, seasonal variables 00:08:58.760 --> 00:09:01.750 align:start position:0% contributions. On the other hand, seasonal variables showed <00:08:59.179><c>negative </c><00:08:59.598><c>importance, </c><00:09:00.017><c>suggesting </c><00:09:00.436><c>that </c><00:09:00.855><c>the </c><00:09:01.274><c>model</c> 00:09:01.750 --> 00:09:01.760 align:start position:0% showed negative importance, suggesting that the model 00:09:01.760 --> 00:09:03.990 align:start position:0% showed negative importance, suggesting that the model deemed <00:09:02.519><c>seasonality </c><00:09:03.278><c>unimportant.</c> 00:09:03.990 --> 00:09:04.000 align:start position:0% deemed seasonality unimportant. 00:09:04.000 --> 00:09:07.069 align:start position:0% deemed seasonality unimportant. In <00:09:04.700><c>other </c><00:09:05.400><c>words, </c><00:09:06.100><c>the </c><00:09:06.800><c>most</c> 00:09:07.069 --> 00:09:07.079 align:start position:0% In other words, the most 00:09:07.079 --> 00:09:10.150 align:start position:0% In other words, the most important <00:09:07.212><c>variable </c><00:09:07.345><c>in </c><00:09:07.478><c>sales </c><00:09:07.611><c>volume </c><00:09:07.744><c>forecasting </c><00:09:07.877><c>is </c><00:09:08.010><c>It </c><00:09:08.143><c>can </c><00:09:08.276><c>be </c><00:09:08.409><c>seen </c><00:09:08.542><c>that </c><00:09:08.675><c>these </c><00:09:08.808><c>factors, </c><00:09:08.941><c>such </c><00:09:09.074><c>as </c><00:09:09.207><c>events </c><00:09:09.340><c>and </c><00:09:09.473><c>prices,</c> 00:09:10.150 --> 00:09:10.160 align:start position:0% important variable in sales volume forecasting is It can be seen that these factors, such as events and prices, 00:09:10.160 --> 00:09:12.670 align:start position:0% important variable in sales volume forecasting is It can be seen that these factors, such as events and prices, have <00:09:10.486><c>a </c><00:09:10.812><c>significant </c><00:09:11.138><c>impact </c><00:09:11.464><c>on </c><00:09:11.790><c>sales. </c><00:09:12.116><c>Factors</c> 00:09:12.670 --> 00:09:15.389 align:start position:0% have a significant impact on sales. Factors 00:09:15.389 --> 00:09:15.399 align:start position:0% 00:09:15.399 --> 00:09:17.910 align:start position:0% such <00:09:15.696><c>as </c><00:09:15.993><c>lead </c><00:09:16.290><c>time, </c><00:09:16.587><c>event </c><00:09:16.884><c>availability, </c><00:09:17.181><c>and </c><00:09:17.478><c>price</c> 00:09:17.910 --> 00:09:17.920 align:start position:0% such as lead time, event availability, and price 00:09:17.920 --> 00:09:19.949 align:start position:0% such as lead time, event availability, and price have <00:09:18.111><c>a </c><00:09:18.302><c>significant </c><00:09:18.493><c>influence </c><00:09:18.684><c>on </c><00:09:18.875><c>model </c><00:09:19.066><c>forecasting; </c><00:09:19.257><c>in </c><00:09:19.448><c>particular, </c><00:09:19.639><c>lead</c> 00:09:19.949 --> 00:09:19.959 align:start position:0% have a significant influence on model forecasting; in particular, lead 00:09:19.959 --> 00:09:22.230 align:start position:0% have a significant influence on model forecasting; in particular, lead time <00:09:21.760><c>is</c> 00:09:22.230 --> 00:09:22.240 align:start position:0% time is 00:09:22.240 --> 00:09:25.790 align:start position:0% time is important <00:09:22.436><c>for </c><00:09:22.632><c>inventory </c><00:09:22.828><c>management </c><00:09:23.024><c>and </c><00:09:23.220><c>delivery </c><00:09:23.416><c>strategies </c><00:09:23.612><c>as </c><00:09:23.808><c>it </c><00:09:24.004><c>represents </c><00:09:24.200><c>the </c><00:09:24.396><c>supply </c><00:09:24.592><c>preparation </c><00:09:24.788><c>period.</c> 00:09:25.790 --> 00:09:25.800 align:start position:0% important for inventory management and delivery strategies as it represents the supply preparation period. 00:09:25.800 --> 00:09:28.110 align:start position:0% important for inventory management and delivery strategies as it represents the supply preparation period. Seasonal <00:09:26.040><c>variables </c><00:09:26.280><c>can </c><00:09:26.520><c>actually </c><00:09:26.760><c>degrade </c><00:09:27.000><c>model </c><00:09:27.240><c>performance, </c><00:09:27.480><c>so </c><00:09:27.720><c>they</c> 00:09:28.110 --> 00:09:29.870 align:start position:0% Seasonal variables can actually degrade model performance, so they 00:09:29.870 --> 00:09:32.670 align:start position:0% 00:09:32.670 --> 00:09:32.680 align:start position:0% 00:09:32.680 --> 00:09:36.069 align:start position:0% should <00:09:33.000><c>be </c><00:09:33.320><c>removed </c><00:09:33.640><c>or </c><00:09:33.960><c>handled </c><00:09:34.280><c>in </c><00:09:34.600><c>a </c><00:09:34.920><c>different </c><00:09:35.240><c>way.</c> 00:09:36.069 --> 00:09:36.079 align:start position:0% should be removed or handled in a different way. 00:09:36.079 --> 00:09:38.710 align:start position:0% should be removed or handled in a different way. Variables <00:09:36.275><c>showing </c><00:09:36.471><c>low </c><00:09:36.667><c>importance, </c><00:09:36.863><c>such </c><00:09:37.059><c>as </c><00:09:37.255><c>holidays, </c><00:09:37.451><c>weekend </c><00:09:37.647><c>availability, </c><00:09:37.843><c>and </c><00:09:38.039><c>weather,</c> 00:09:38.710 --> 00:09:38.720 align:start position:0% Variables showing low importance, such as holidays, weekend availability, and weather, 00:09:38.720 --> 00:09:40.670 align:start position:0% Variables showing low importance, such as holidays, weekend availability, and weather, may <00:09:39.019><c>be </c><00:09:39.318><c>significant </c><00:09:39.617><c>within </c><00:09:39.916><c>specific </c><00:09:40.215><c>categories, </c><00:09:40.514><c>so</c> 00:09:40.670 --> 00:09:40.680 align:start position:0% may be significant within specific categories, so 00:09:40.680 --> 00:09:43.670 align:start position:0% may be significant within specific categories, so granular <00:09:41.346><c>analysis </c><00:09:42.012><c>is </c><00:09:42.678><c>required.</c> 00:09:43.670 --> 00:09:46.269 align:start position:0% granular analysis is required. 00:09:46.269 --> 00:09:46.279 align:start position:0% 00:09:46.279 --> 00:09:50.110 align:start position:0% The <00:09:46.439><c>variables </c><00:09:46.599><c>that </c><00:09:46.759><c>have </c><00:09:46.919><c>the </c><00:09:47.079><c>greatest </c><00:09:47.239><c>impact </c><00:09:47.399><c>on </c><00:09:47.559><c>sales </c><00:09:47.719><c>volume </c><00:09:47.879><c>forecasting </c><00:09:48.039><c>are </c><00:09:48.199><c>inventory </c><00:09:48.359><c>quantity, </c><00:09:48.519><c>price, </c><00:09:48.679><c>and </c><00:09:48.839><c>competitor </c><00:09:48.999><c>price, </c><00:09:49.159><c>and</c> 00:09:50.110 --> 00:09:50.120 align:start position:0% The variables that have the greatest impact on sales volume forecasting are inventory quantity, price, and competitor price, and 00:09:50.120 --> 00:09:52.829 align:start position:0% The variables that have the greatest impact on sales volume forecasting are inventory quantity, price, and competitor price, and these <00:09:50.344><c>have </c><00:09:50.568><c>a </c><00:09:50.792><c>positive </c><00:09:51.016><c>correlation </c><00:09:51.240><c>with </c><00:09:51.464><c>sales </c><00:09:51.688><c>volume. </c><00:09:51.912><c>Factors </c><00:09:52.136><c>such </c><00:09:52.360><c>as</c> 00:09:52.829 --> 00:09:52.839 align:start position:0% these have a positive correlation with sales volume. Factors such as 00:09:52.839 --> 00:09:56.269 align:start position:0% these have a positive correlation with sales volume. Factors such as season, <00:09:53.391><c>holidays, </c><00:09:53.943><c>and </c><00:09:54.495><c>weekend </c><00:09:55.047><c>availability </c><00:09:55.599><c>have</c> 00:09:56.269 --> 00:09:58.630 align:start position:0% season, holidays, and weekend availability have 00:09:58.630 --> 00:09:58.640 align:start position:0% 00:09:58.640 --> 00:10:01.670 align:start position:0% low <00:09:58.813><c>influence </c><00:09:58.986><c>based </c><00:09:59.159><c>on </c><00:09:59.332><c>linear </c><00:09:59.505><c>correlation, </c><00:09:59.678><c>but </c><00:09:59.851><c>they </c><00:10:00.024><c>can </c><00:10:00.197><c>interact </c><00:10:00.370><c>in </c><00:10:00.543><c>non-linear </c><00:10:00.716><c>models, </c><00:10:00.889><c>so </c><00:10:01.062><c>their </c><00:10:01.235><c>removal</c> 00:10:01.670 --> 00:10:01.680 align:start position:0% low influence based on linear correlation, but they can interact in non-linear models, so their removal 00:10:01.680 --> 00:10:04.550 align:start position:0% low influence based on linear correlation, but they can interact in non-linear models, so their removal must <00:10:02.170><c>be </c><00:10:02.660><c>done </c><00:10:03.150><c>with </c><00:10:03.640><c>caution.</c> 00:10:04.550 --> 00:10:04.560 align:start position:0% must be done with caution. 00:10:04.560 --> 00:10:07.230 align:start position:0% must be done with caution. To <00:10:04.885><c>improve </c><00:10:05.210><c>model </c><00:10:05.535><c>performance, </c><00:10:05.860><c>non-linear </c><00:10:06.185><c>relationships </c><00:10:06.510><c>must </c><00:10:06.835><c>be</c> 00:10:07.230 --> 00:10:07.240 align:start position:0% To improve model performance, non-linear relationships must be 00:10:07.240 --> 00:10:10.150 align:start position:0% To improve model performance, non-linear relationships must be considered, <00:10:07.524><c>relationships </c><00:10:07.808><c>between </c><00:10:08.092><c>variables </c><00:10:08.376><c>analyzed </c><00:10:08.660><c>using </c><00:10:08.944><c>various </c><00:10:09.228><c>models, </c><00:10:09.512><c>and </c><00:10:09.796><c>domain</c> 00:10:10.150 --> 00:10:10.160 align:start position:0% considered, relationships between variables analyzed using various models, and domain 00:10:10.160 --> 00:10:12.990 align:start position:0% considered, relationships between variables analyzed using various models, and domain knowledge 00:10:12.990 --> 00:10:13.000 align:start position:0% knowledge 00:10:13.000 --> 00:10:15.670 align:start position:0% knowledge utilized. <00:10:13.159><c>There </c><00:10:13.318><c>is </c><00:10:13.477><c>no </c><00:10:13.636><c>distinct </c><00:10:13.795><c>linear </c><00:10:13.954><c>relationship </c><00:10:14.113><c>between </c><00:10:14.272><c>mid-time </c><00:10:14.431><c>and </c><00:10:14.590><c>sales </c><00:10:14.749><c>volume, </c><00:10:14.908><c>and </c><00:10:15.067><c>the</c> 00:10:15.670 --> 00:10:18.310 align:start position:0% utilized. There is no distinct linear relationship between mid-time and sales volume, and the 00:10:18.310 --> 00:10:18.320 align:start position:0% 00:10:18.320 --> 00:10:21.069 align:start position:0% correlation <00:10:18.688><c>between </c><00:10:19.056><c>the </c><00:10:19.424><c>two </c><00:10:19.792><c>variables </c><00:10:20.160><c>is</c> 00:10:21.069 --> 00:10:21.079 align:start position:0% correlation between the two variables is 00:10:21.079 --> 00:10:24.430 align:start position:0% correlation between the two variables is very <00:10:21.669><c>low. </c><00:10:22.259><c>Some </c><00:10:22.849><c>data </c><00:10:23.439><c>points</c> 00:10:24.430 --> 00:10:24.440 align:start position:0% very low. Some data points 00:10:24.440 --> 00:10:26.990 align:start position:0% very low. Some data points deviate <00:10:24.920><c>from </c><00:10:25.400><c>the </c><00:10:25.880><c>overall </c><00:10:26.360><c>distribution</c> 00:10:26.990 --> 00:10:27.000 align:start position:0% deviate from the overall distribution 00:10:27.000 --> 00:10:29.630 align:start position:0% deviate from the overall distribution and <00:10:27.464><c>may </c><00:10:27.928><c>be </c><00:10:28.392><c>considered </c><00:10:28.856><c>outliers, </c><00:10:29.320><c>so</c> 00:10:29.630 --> 00:10:29.640 align:start position:0% and may be considered outliers, so 00:10:29.640 --> 00:10:32.150 align:start position:0% and may be considered outliers, so caution <00:10:29.847><c>is </c><00:10:30.054><c>required </c><00:10:30.261><c>during </c><00:10:30.468><c>data </c><00:10:30.675><c>analysis. </c><00:10:30.882><c>Since </c><00:10:31.089><c>factors </c><00:10:31.296><c>other </c><00:10:31.503><c>than </c><00:10:31.710><c>lead </c><00:10:31.917><c>time</c> 00:10:32.150 --> 00:10:32.160 align:start position:0% caution is required during data analysis. Since factors other than lead time 00:10:32.160 --> 00:10:34.670 align:start position:0% caution is required during data analysis. Since factors other than lead time can <00:10:32.933><c>affect </c><00:10:33.706><c>sales </c><00:10:34.479><c>volume,</c> 00:10:34.670 --> 00:10:34.680 align:start position:0% can affect sales volume, 00:10:34.680 --> 00:10:37.550 align:start position:0% can affect sales volume, additional <00:10:35.479><c>analysis </c><00:10:36.278><c>and </c><00:10:37.077><c>data</c> 00:10:37.550 --> 00:10:37.560 align:start position:0% additional analysis and data 00:10:37.560 --> 00:10:40.310 align:start position:0% additional analysis and data collection <00:10:37.893><c>are </c><00:10:38.226><c>necessary. </c><00:10:38.559><c>The </c><00:10:38.892><c>TFT </c><00:10:39.225><c>model </c><00:10:39.558><c>shows</c> 00:10:40.310 --> 00:10:42.870 align:start position:0% collection are necessary. The TFT model shows 00:10:42.870 --> 00:10:42.880 align:start position:0% 00:10:42.880 --> 00:10:45.150 align:start position:0% nearly <00:10:42.960><c>equal </c><00:10:43.040><c>importance </c><00:10:43.120><c>to </c><00:10:43.200><c>the </c><00:10:43.280><c>workforce </c><00:10:43.360><c>variable </c><00:10:43.440><c>at </c><00:10:43.520><c>the </c><00:10:43.600><c>last </c><00:10:43.680><c>encoder </c><00:10:43.760><c>point. </c><00:10:43.840><c>It </c><00:10:43.920><c>assigns </c><00:10:44.000><c>weights, </c><00:10:44.080><c>meaning </c><00:10:44.160><c>the</c> 00:10:45.150 --> 00:10:45.160 align:start position:0% nearly equal importance to the workforce variable at the last encoder point. It assigns weights, meaning the 00:10:45.160 --> 00:10:48.230 align:start position:0% nearly equal importance to the workforce variable at the last encoder point. It assigns weights, meaning the model <00:10:45.350><c>makes </c><00:10:45.540><c>judgments </c><00:10:45.730><c>by </c><00:10:45.920><c>combining </c><00:10:46.110><c>various </c><00:10:46.300><c>variables </c><00:10:46.490><c>rather </c><00:10:46.680><c>than </c><00:10:46.870><c>relying </c><00:10:47.060><c>excessively </c><00:10:47.250><c>on </c><00:10:47.440><c>specific </c><00:10:47.630><c>variables.</c> 00:10:48.230 --> 00:10:51.550 align:start position:0% model makes judgments by combining various variables rather than relying excessively on specific variables. 00:10:51.550 --> 00:10:51.560 align:start position:0% 00:10:51.560 --> 00:10:54.949 align:start position:0% The <00:10:52.310><c>weights </c><00:10:53.060><c>of </c><00:10:53.810><c>all </c><00:10:54.560><c>variables</c> 00:10:54.949 --> 00:10:54.959 align:start position:0% The weights of all variables 00:10:54.959 --> 00:10:57.350 align:start position:0% The weights of all variables are <00:10:55.066><c>similar </c><00:10:55.173><c>at </c><00:10:55.280><c>a </c><00:10:55.387><c>level </c><00:10:55.494><c>of </c><00:10:55.601><c>approximately </c><00:10:55.708><c>0.1, </c><00:10:55.815><c>showing </c><00:10:55.922><c>a </c><00:10:56.029><c>pattern </c><00:10:56.136><c>different </c><00:10:56.243><c>from </c><00:10:56.350><c>the </c><00:10:56.457><c>general </c><00:10:56.564><c>assumption </c><00:10:56.671><c>that</c> 00:10:57.350 --> 00:10:57.360 align:start position:0% are similar at a level of approximately 0.1, showing a pattern different from the general assumption that 00:10:57.360 --> 00:10:59.310 align:start position:0% are similar at a level of approximately 0.1, showing a pattern different from the general assumption that high <00:10:57.554><c>selection </c><00:10:57.748><c>weights </c><00:10:57.942><c>are </c><00:10:58.136><c>assigned </c><00:10:58.330><c>to </c><00:10:58.524><c>specific </c><00:10:58.718><c>variables.</c> 00:10:59.310 --> 00:11:02.350 align:start position:0% high selection weights are assigned to specific variables. 00:11:02.350 --> 00:11:02.360 align:start position:0% 00:11:02.360 --> 00:11:04.829 align:start position:0% Data <00:11:02.880><c>normalization </c><00:11:03.400><c>or </c><00:11:03.920><c>attention </c><00:11:04.440><c>layer</c> 00:11:04.829 --> 00:11:04.839 align:start position:0% Data normalization or attention layer 00:11:04.839 --> 00:11:07.230 align:start position:0% Data normalization or attention layer tuning <00:11:05.039><c>may </c><00:11:05.239><c>still </c><00:11:05.439><c>be </c><00:11:05.639><c>in </c><00:11:05.839><c>the </c><00:11:06.039><c>early </c><00:11:06.239><c>stages; </c><00:11:06.439><c>while </c><00:11:06.639><c>the</c> 00:11:07.230 --> 00:11:07.240 align:start position:0% tuning may still be in the early stages; while the 00:11:07.240 --> 00:11:09.670 align:start position:0% tuning may still be in the early stages; while the model's <00:11:07.749><c>generalization </c><00:11:08.258><c>performance </c><00:11:08.767><c>is </c><00:11:09.276><c>high,</c> 00:11:09.670 --> 00:11:09.680 align:start position:0% model's generalization performance is high, 00:11:09.680 --> 00:11:12.430 align:start position:0% model's generalization performance is high, model <00:11:10.020><c>improvement </c><00:11:10.360><c>may </c><00:11:10.700><c>be </c><00:11:11.040><c>considered </c><00:11:11.380><c>as </c><00:11:11.720><c>needed.</c> 00:11:12.430 --> 00:11:12.440 align:start position:0% model improvement may be considered as needed. 00:11:12.440 --> 00:11:15.269 align:start position:0% model improvement may be considered as needed. Variable <00:11:13.580><c>Selection </c><00:11:14.720><c>Ways</c> 00:11:15.269 --> 00:11:15.279 align:start position:0% Variable Selection Ways 00:11:15.279 --> 00:11:17.230 align:start position:0% Variable Selection Ways assign <00:11:15.519><c>equal </c><00:11:15.759><c>importance </c><00:11:15.999><c>to </c><00:11:16.239><c>all </c><00:11:16.479><c>variables, </c><00:11:16.719><c>whereas</c> 00:11:17.230 --> 00:11:17.240 align:start position:0% assign equal importance to all variables, whereas 00:11:17.240 --> 00:11:18.389 align:start position:0% assign equal importance to all variables, whereas Permutation 00:11:18.389 --> 00:11:18.399 align:start position:0% Permutation 00:11:18.399 --> 00:11:21.310 align:start position:0% Permutation Forces <00:11:18.689><c>show </c><00:11:18.979><c>a </c><00:11:19.269><c>significant </c><00:11:19.559><c>difference </c><00:11:19.849><c>in </c><00:11:20.139><c>importance </c><00:11:20.429><c>per </c><00:11:20.719><c>variable.</c> 00:11:21.310 --> 00:11:21.320 align:start position:0% Forces show a significant difference in importance per variable. 00:11:21.320 --> 00:11:24.230 align:start position:0% Forces show a significant difference in importance per variable. Variable <00:11:22.340><c>Selection </c><00:11:23.360><c>Ways</c> 00:11:24.230 --> 00:11:26.190 align:start position:0% Variable Selection Ways 00:11:26.190 --> 00:11:26.200 align:start position:0% 00:11:26.200 --> 00:11:28.190 align:start position:0% indicate <00:11:26.352><c>that </c><00:11:26.504><c>the </c><00:11:26.656><c>model </c><00:11:26.808><c>predicts </c><00:11:26.960><c>by </c><00:11:27.112><c>combining </c><00:11:27.264><c>various </c><00:11:27.416><c>variables, </c><00:11:27.568><c>while </c><00:11:27.720><c>Permutation</c> 00:11:28.190 --> 00:11:28.200 align:start position:0% indicate that the model predicts by combining various variables, while Permutation 00:11:28.200 --> 00:11:31.389 align:start position:0% indicate that the model predicts by combining various variables, while Permutation Forces <00:11:28.529><c>indicate </c><00:11:28.858><c>that </c><00:11:29.187><c>the </c><00:11:29.516><c>model </c><00:11:29.845><c>relies </c><00:11:30.174><c>on </c><00:11:30.503><c>specific </c><00:11:30.832><c>variables.</c> 00:11:31.389 --> 00:11:31.399 align:start position:0% Forces indicate that the model relies on specific variables. 00:11:31.399 --> 00:11:33.949 align:start position:0% Forces indicate that the model relies on specific variables. When <00:11:31.999><c>developing </c><00:11:32.599><c>a </c><00:11:33.199><c>model, </c><00:11:33.799><c>both</c> 00:11:33.949 --> 00:11:33.959 align:start position:0% When developing a model, both 00:11:33.959 --> 00:11:36.910 align:start position:0% When developing a model, both methods <00:11:34.439><c>should </c><00:11:34.919><c>be </c><00:11:35.399><c>used </c><00:11:35.879><c>together </c><00:11:36.359><c>to</c> 00:11:36.910 --> 00:11:36.920 align:start position:0% methods should be used together to 00:11:36.920 --> 00:11:39.389 align:start position:0% methods should be used together to comprehensively <00:11:37.150><c>assess </c><00:11:37.380><c>variable </c><00:11:37.610><c>importance </c><00:11:37.840><c>and </c><00:11:38.070><c>improve </c><00:11:38.300><c>model </c><00:11:38.530><c>performance. </c><00:11:38.760><c>This</c> 00:11:39.389 --> 00:11:39.399 align:start position:0% comprehensively assess variable importance and improve model performance. This 00:11:39.399 --> 00:11:42.150 align:start position:0% comprehensively assess variable importance and improve model performance. This graph 00:11:42.150 --> 00:11:45.470 align:start position:0% graph 00:11:45.470 --> 00:11:47.590 align:start position:0% 00:11:47.590 --> 00:11:47.600 align:start position:0% 00:11:47.600 --> 00:11:50.430 align:start position:0% visually 00:11:50.430 --> 00:11:50.440 align:start position:0% visually 00:11:50.440 --> 00:11:52.829 align:start position:0% visually displays <00:11:50.527><c>model </c><00:11:50.614><c>performance </c><00:11:50.701><c>by </c><00:11:50.788><c>comparing </c><00:11:50.875><c>the </c><00:11:50.962><c>prediction </c><00:11:51.049><c>results </c><00:11:51.136><c>of </c><00:11:51.223><c>various </c><00:11:51.310><c>models, </c><00:11:51.397><c>such </c><00:11:51.484><c>as </c><00:11:51.571><c>Random </c><00:11:51.658><c>Forest, </c><00:11:51.745><c>TFT, </c><00:11:51.832><c>and </c><00:11:51.919><c>Ensemble, </c><00:11:52.006><c>with </c><00:11:52.093><c>actual </c><00:11:52.180><c>values. </c><00:11:52.267><c>By </c><00:11:52.354><c>comparing</c> 00:11:52.829 --> 00:11:52.839 align:start position:0% displays model performance by comparing the prediction results of various models, such as Random Forest, TFT, and Ensemble, with actual values. By comparing 00:11:52.839 --> 00:11:55.949 align:start position:0% displays model performance by comparing the prediction results of various models, such as Random Forest, TFT, and Ensemble, with actual values. By comparing the <00:11:53.189><c>predicted </c><00:11:53.539><c>values </c><00:11:53.889><c>of </c><00:11:54.239><c>each </c><00:11:54.589><c>model </c><00:11:54.939><c>with </c><00:11:55.289><c>the </c><00:11:55.639><c>actual</c> 00:11:55.949 --> 00:11:55.959 align:start position:0% the predicted values of each model with the actual 00:11:55.959 --> 00:11:58.230 align:start position:0% the predicted values of each model with the actual values <00:11:56.351><c>over </c><00:11:56.743><c>50 </c><00:11:57.135><c>data </c><00:11:57.527><c>points, </c><00:11:57.919><c>you</c> 00:11:58.230 --> 00:11:58.240 align:start position:0% values over 50 data points, you 00:11:58.240 --> 00:12:00.550 align:start position:0% values over 50 data points, you can <00:11:58.509><c>identify </c><00:11:58.778><c>which </c><00:11:59.047><c>model </c><00:11:59.316><c>provides </c><00:11:59.585><c>the </c><00:11:59.854><c>most </c><00:12:00.123><c>accurate </c><00:12:00.392><c>prediction.</c> 00:12:00.550 --> 00:12:00.560 align:start position:0% can identify which model provides the most accurate prediction. 00:12:00.560 --> 00:12:03.949 align:start position:0% can identify which model provides the most accurate prediction. The <00:12:01.960><c>graph </c><00:12:03.360><c>helps</c> 00:12:03.949 --> 00:12:03.959 align:start position:0% The graph helps 00:12:03.959 --> 00:12:06.350 align:start position:0% The graph helps analyze <00:12:04.115><c>errors </c><00:12:04.271><c>by </c><00:12:04.427><c>model </c><00:12:04.583><c>and </c><00:12:04.739><c>assists </c><00:12:04.895><c>in </c><00:12:05.051><c>selecting </c><00:12:05.207><c>the </c><00:12:05.363><c>most </c><00:12:05.519><c>suitable </c><00:12:05.675><c>model.</c> 00:12:06.350 --> 00:12:09.030 align:start position:0% analyze errors by model and assists in selecting the most suitable model. 00:12:09.030 --> 00:12:09.040 align:start position:0% 00:12:09.040 --> 00:12:11.470 align:start position:0% Overall, <00:12:10.100><c>Ensemble </c><00:12:11.160><c>models</c> 00:12:11.470 --> 00:12:11.480 align:start position:0% Overall, Ensemble models 00:12:11.480 --> 00:12:13.550 align:start position:0% Overall, Ensemble models predict <00:12:11.700><c>sales </c><00:12:11.920><c>volume </c><00:12:12.140><c>lower </c><00:12:12.360><c>than </c><00:12:12.580><c>actual </c><00:12:12.800><c>sales </c><00:12:13.020><c>volume. </c><00:12:13.240><c>It</c> 00:12:13.550 --> 00:12:13.560 align:start position:0% predict sales volume lower than actual sales volume. It 00:12:13.560 --> 00:12:16.430 align:start position:0% predict sales volume lower than actual sales volume. It exhibits <00:12:13.927><c>a </c><00:12:14.294><c>predictive </c><00:12:14.661><c>tendency, </c><00:12:15.028><c>with </c><00:12:15.395><c>a</c> 00:12:16.430 --> 00:12:16.440 align:start position:0% exhibits a predictive tendency, with a 00:12:16.440 --> 00:12:20.150 align:start position:0% exhibits a predictive tendency, with a significant <00:12:18.080><c>prediction </c><00:12:19.720><c>error</c> 00:12:20.150 --> 00:12:20.160 align:start position:0% significant prediction error 00:12:20.160 --> 00:12:22.550 align:start position:0% significant prediction error appearing <00:12:20.273><c>particularly </c><00:12:20.386><c>on </c><00:12:20.499><c>days </c><00:12:20.612><c>4, </c><00:12:20.725><c>7, </c><00:12:20.838><c>and </c><00:12:20.951><c>8. </c><00:12:21.064><c>Problems </c><00:12:21.177><c>arise </c><00:12:21.290><c>where </c><00:12:21.403><c>prediction </c><00:12:21.516><c>sensitivity </c><00:12:21.629><c>is </c><00:12:21.742><c>insufficient </c><00:12:21.855><c>in </c><00:12:21.968><c>high-demand </c><00:12:22.081><c>periods, </c><00:12:22.194><c>and</c> 00:12:22.550 --> 00:12:25.069 align:start position:0% appearing particularly on days 4, 7, and 8. Problems arise where prediction sensitivity is insufficient in high-demand periods, and 00:12:25.069 --> 00:12:27.629 align:start position:0% 00:12:27.629 --> 00:12:27.639 align:start position:0% 00:12:27.639 --> 00:12:30.069 align:start position:0% prediction <00:12:27.769><c>accuracy </c><00:12:27.899><c>drops </c><00:12:28.029><c>in </c><00:12:28.159><c>periods </c><00:12:28.289><c>of </c><00:12:28.419><c>large </c><00:12:28.549><c>fluctuations, </c><00:12:28.679><c>such </c><00:12:28.809><c>as </c><00:12:28.939><c>immediately </c><00:12:29.069><c>after </c><00:12:29.199><c>marketing </c><00:12:29.329><c>events </c><00:12:29.459><c>or </c><00:12:29.589><c>restocking.</c> 00:12:30.069 --> 00:12:32.829 align:start position:0% prediction accuracy drops in periods of large fluctuations, such as immediately after marketing events or restocking. 00:12:32.829 --> 00:12:34.750 align:start position:0% 00:12:34.750 --> 00:12:34.760 align:start position:0% 00:12:34.760 --> 00:12:37.590 align:start position:0% Particular <00:12:34.899><c>caution </c><00:12:35.038><c>is </c><00:12:35.177><c>required </c><00:12:35.316><c>in </c><00:12:35.455><c>periods </c><00:12:35.594><c>where </c><00:12:35.733><c>demand </c><00:12:35.872><c>forecasting </c><00:12:36.011><c>failure </c><00:12:36.150><c>can </c><00:12:36.289><c>have </c><00:12:36.428><c>a </c><00:12:36.567><c>significant </c><00:12:36.706><c>impact </c><00:12:36.845><c>on </c><00:12:36.984><c>the </c><00:12:37.123><c>business, </c><00:12:37.262><c>and</c> 00:12:37.590 --> 00:12:37.600 align:start position:0% Particular caution is required in periods where demand forecasting failure can have a significant impact on the business, and 00:12:37.600 --> 00:12:40.670 align:start position:0% Particular caution is required in periods where demand forecasting failure can have a significant impact on the business, and separate 00:12:40.670 --> 00:12:40.680 align:start position:0% separate 00:12:40.680 --> 00:12:43.350 align:start position:0% separate modeling <00:12:40.863><c>strategies </c><00:12:41.046><c>for </c><00:12:41.229><c>high </c><00:12:41.412><c>volatility </c><00:12:41.595><c>should </c><00:12:41.778><c>be </c><00:12:41.961><c>considered </c><00:12:42.144><c>alongside </c><00:12:42.327><c>model </c><00:12:42.510><c>improvements. </c><00:12:42.693><c>The </c><00:12:42.876><c>TFT</c> 00:12:43.350 --> 00:12:43.360 align:start position:0% modeling strategies for high volatility should be considered alongside model improvements. The TFT 00:12:43.360 --> 00:12:46.110 align:start position:0% modeling strategies for high volatility should be considered alongside model improvements. The TFT model <00:12:43.520><c>is </c><00:12:43.680><c>a </c><00:12:43.840><c>deep </c><00:12:44.000><c>learning-based </c><00:12:44.160><c>model </c><00:12:44.320><c>that </c><00:12:44.480><c>predicts </c><00:12:44.640><c>future </c><00:12:44.800><c>demand </c><00:12:44.960><c>by </c><00:12:45.120><c>learning </c><00:12:45.280><c>from </c><00:12:45.440><c>time-series </c><00:12:45.600><c>data. </c><00:12:45.760><c>It</c> 00:12:46.110 --> 00:12:48.430 align:start position:0% model is a deep learning-based model that predicts future demand by learning from time-series data. It 00:12:48.430 --> 00:12:51.230 align:start position:0% 00:12:51.230 --> 00:12:53.269 align:start position:0% 00:12:53.269 --> 00:12:53.279 align:start position:0% 00:12:53.279 --> 00:12:55.470 align:start position:0% has <00:12:53.442><c>a </c><00:12:53.605><c>structure </c><00:12:53.768><c>capable </c><00:12:53.931><c>of </c><00:12:54.094><c>effectively </c><00:12:54.257><c>forecasting </c><00:12:54.420><c>demand </c><00:12:54.583><c>by </c><00:12:54.746><c>considering </c><00:12:54.909><c>multivariate </c><00:12:55.072><c>characteristics.</c> 00:12:55.470 --> 00:12:55.480 align:start position:0% has a structure capable of effectively forecasting demand by considering multivariate characteristics. 00:12:55.480 --> 00:12:57.990 align:start position:0% has a structure capable of effectively forecasting demand by considering multivariate characteristics. In <00:12:55.636><c>this </c><00:12:55.792><c>study, </c><00:12:55.948><c>the </c><00:12:56.104><c>model </c><00:12:56.260><c>was </c><00:12:56.416><c>trained </c><00:12:56.572><c>using </c><00:12:56.728><c>data </c><00:12:56.884><c>from </c><00:12:57.040><c>the </c><00:12:57.196><c>past </c><00:12:57.352><c>30 </c><00:12:57.508><c>days</c> 00:12:57.990 --> 00:12:58.000 align:start position:0% In this study, the model was trained using data from the past 30 days 00:12:58.000 --> 00:13:00.550 align:start position:0% In this study, the model was trained using data from the past 30 days as <00:12:58.212><c>the </c><00:12:58.424><c>encoder </c><00:12:58.636><c>and </c><00:12:58.848><c>a </c><00:12:59.060><c>forecast </c><00:12:59.272><c>period </c><00:12:59.484><c>of </c><00:12:59.696><c>7 </c><00:12:59.908><c>days. </c><00:13:00.120><c>This</c> 00:13:00.550 --> 00:13:02.750 align:start position:0% as the encoder and a forecast period of 7 days. This 00:13:02.750 --> 00:13:02.760 align:start position:0% 00:13:02.760 --> 00:13:04.910 align:start position:0% graph <00:13:03.560><c>compares </c><00:13:04.360><c>the</c> 00:13:04.910 --> 00:13:04.920 align:start position:0% graph compares the 00:13:04.920 --> 00:13:07.670 align:start position:0% graph compares the prediction <00:13:05.191><c>results </c><00:13:05.462><c>of </c><00:13:05.733><c>the </c><00:13:06.004><c>TFT </c><00:13:06.275><c>model </c><00:13:06.546><c>and </c><00:13:06.817><c>the </c><00:13:07.088><c>RF </c><00:13:07.359><c>model</c> 00:13:07.670 --> 00:13:07.680 align:start position:0% prediction results of the TFT model and the RF model 00:13:07.680 --> 00:13:10.790 align:start position:0% prediction results of the TFT model and the RF model with <00:13:08.920><c>actual </c><00:13:10.160><c>values.</c> 00:13:10.790 --> 00:13:10.800 align:start position:0% with actual values. 00:13:10.800 --> 00:13:12.910 align:start position:0% with actual values. By 00:13:12.910 --> 00:13:12.920 align:start position:0% By 00:13:12.920 --> 00:13:15.629 align:start position:0% By visualizing <00:13:13.012><c>the </c><00:13:13.104><c>30 </c><00:13:13.196><c>samples </c><00:13:13.288><c>prior </c><00:13:13.380><c>to </c><00:13:13.472><c>the </c><00:13:13.564><c>day </c><00:13:13.656><c>7 </c><00:13:13.748><c>prediction </c><00:13:13.840><c>value, </c><00:13:13.932><c>it </c><00:13:14.024><c>shows </c><00:13:14.116><c>how </c><00:13:14.208><c>closely </c><00:13:14.300><c>each </c><00:13:14.392><c>model </c><00:13:14.484><c>matches </c><00:13:14.576><c>the </c><00:13:14.668><c>actual </c><00:13:14.760><c>values.</c> 00:13:15.629 --> 00:13:18.430 align:start position:0% visualizing the 30 samples prior to the day 7 prediction value, it shows how closely each model matches the actual values. 00:13:18.430 --> 00:13:18.440 align:start position:0% 00:13:18.440 --> 00:13:21.470 align:start position:0% By <00:13:18.702><c>comparing </c><00:13:18.964><c>the </c><00:13:19.226><c>prediction </c><00:13:19.488><c>patterns </c><00:13:19.750><c>of </c><00:13:20.012><c>the </c><00:13:20.274><c>two </c><00:13:20.536><c>models, </c><00:13:20.798><c>the</c> 00:13:21.470 --> 00:13:21.480 align:start position:0% By comparing the prediction patterns of the two models, the 00:13:21.480 --> 00:13:24.350 align:start position:0% By comparing the prediction patterns of the two models, the performance <00:13:21.724><c>superiority </c><00:13:21.968><c>of </c><00:13:22.212><c>the </c><00:13:22.456><c>TFT </c><00:13:22.700><c>model </c><00:13:22.944><c>can </c><00:13:23.188><c>be </c><00:13:23.432><c>visually </c><00:13:23.676><c>evaluated. </c><00:13:23.920><c>This</c> 00:13:24.350 --> 00:13:24.360 align:start position:0% performance superiority of the TFT model can be visually evaluated. This 00:13:24.360 --> 00:13:27.269 align:start position:0% performance superiority of the TFT model can be visually evaluated. This table <00:13:24.669><c>numerically </c><00:13:24.978><c>represents </c><00:13:25.287><c>the </c><00:13:25.596><c>performance </c><00:13:25.905><c>of </c><00:13:26.214><c>the </c><00:13:26.523><c>TFT </c><00:13:26.832><c>model.</c> 00:13:27.269 --> 00:13:29.750 align:start position:0% table numerically represents the performance of the TFT model. 00:13:29.750 --> 00:13:29.760 align:start position:0% 00:13:29.760 --> 00:13:32.470 align:start position:0% Through <00:13:30.120><c>each </c><00:13:30.480><c>value, </c><00:13:30.840><c>the </c><00:13:31.200><c>model's </c><00:13:31.560><c>prediction </c><00:13:31.920><c>error</c> 00:13:32.470 --> 00:13:32.480 align:start position:0% Through each value, the model's prediction error 00:13:32.480 --> 00:13:35.350 align:start position:0% Through each value, the model's prediction error can <00:13:32.944><c>be </c><00:13:33.408><c>intuitively </c><00:13:33.872><c>verified. </c><00:13:34.336><c>The </c><00:13:34.800><c>TFT</c> 00:13:35.350 --> 00:13:35.360 align:start position:0% can be intuitively verified. The TFT 00:13:35.360 --> 00:13:37.509 align:start position:0% can be intuitively verified. The TFT model's <00:13:35.560><c>error </c><00:13:35.760><c>is</c> 00:13:37.509 --> 00:13:37.519 align:start position:0% model's error is 00:13:37.519 --> 00:13:40.550 align:start position:0% model's error is 29.83%, <00:13:37.959><c>suggesting </c><00:13:38.399><c>potential </c><00:13:38.839><c>for </c><00:13:39.279><c>improving </c><00:13:39.719><c>prediction </c><00:13:40.159><c>accuracy.</c> 00:13:40.550 --> 00:13:40.560 align:start position:0% 29.83%, suggesting potential for improving prediction accuracy. 00:13:40.560 --> 00:13:43.629 align:start position:0% 29.83%, suggesting potential for improving prediction accuracy. The <00:13:43.040><c>graph</c> 00:13:43.629 --> 00:13:46.629 align:start position:0% The graph 00:13:46.629 --> 00:13:46.639 align:start position:0% 00:13:46.639 --> 00:13:49.470 align:start position:0% shows <00:13:46.761><c>the </c><00:13:46.883><c>change </c><00:13:47.005><c>in </c><00:13:47.127><c>MSE </c><00:13:47.249><c>according </c><00:13:47.371><c>to </c><00:13:47.493><c>the </c><00:13:47.615><c>prediction </c><00:13:47.737><c>order. </c><00:13:47.859><c>As </c><00:13:47.981><c>the </c><00:13:48.103><c>prediction </c><00:13:48.225><c>order </c><00:13:48.347><c>increases, </c><00:13:48.469><c>the</c> 00:13:49.470 --> 00:13:49.480 align:start position:0% shows the change in MSE according to the prediction order. As the prediction order increases, the 00:13:49.480 --> 00:13:51.910 align:start position:0% shows the change in MSE according to the prediction order. As the prediction order increases, the MSE 00:13:51.910 --> 00:13:51.920 align:start position:0% MSE 00:13:51.920 --> 00:13:54.030 align:start position:0% MSE shows <00:13:52.165><c>a </c><00:13:52.410><c>pattern </c><00:13:52.655><c>of </c><00:13:52.900><c>gradually </c><00:13:53.145><c>increasing, </c><00:13:53.390><c>reflecting </c><00:13:53.635><c>the</c> 00:13:54.030 --> 00:13:54.040 align:start position:0% shows a pattern of gradually increasing, reflecting the 00:13:54.040 --> 00:13:56.910 align:start position:0% shows a pattern of gradually increasing, reflecting the general <00:13:54.206><c>characteristic </c><00:13:54.372><c>of </c><00:13:54.538><c>time </c><00:13:54.704><c>series </c><00:13:54.870><c>forecasting </c><00:13:55.036><c>where </c><00:13:55.202><c>errors </c><00:13:55.368><c>increase </c><00:13:55.534><c>during </c><00:13:55.700><c>future </c><00:13:55.866><c>predictions. </c><00:13:56.032><c>This</c> 00:13:56.910 --> 00:13:56.920 align:start position:0% general characteristic of time series forecasting where errors increase during future predictions. This 00:13:56.920 --> 00:13:59.470 align:start position:0% general characteristic of time series forecasting where errors increase during future predictions. This graph 00:13:59.470 --> 00:13:59.480 align:start position:0% graph 00:13:59.480 --> 00:14:02.150 align:start position:0% graph compares <00:13:59.629><c>the </c><00:13:59.778><c>average </c><00:13:59.927><c>predicted </c><00:14:00.076><c>value </c><00:14:00.225><c>of </c><00:14:00.374><c>the </c><00:14:00.523><c>entire </c><00:14:00.672><c>sample </c><00:14:00.821><c>with </c><00:14:00.970><c>the </c><00:14:01.119><c>actual </c><00:14:01.268><c>value. </c><00:14:01.417><c>It </c><00:14:01.566><c>allows </c><00:14:01.715><c>for</c> 00:14:02.150 --> 00:14:04.749 align:start position:0% compares the average predicted value of the entire sample with the actual value. It allows for 00:14:04.749 --> 00:14:04.759 align:start position:0% 00:14:04.759 --> 00:14:07.629 align:start position:0% the <00:14:04.873><c>identification </c><00:14:04.987><c>of </c><00:14:05.101><c>the </c><00:14:05.215><c>model's </c><00:14:05.329><c>overall </c><00:14:05.443><c>trend </c><00:14:05.557><c>and </c><00:14:05.671><c>verification </c><00:14:05.785><c>of </c><00:14:05.899><c>how </c><00:14:06.013><c>well </c><00:14:06.127><c>the </c><00:14:06.241><c>model </c><00:14:06.355><c>tracks </c><00:14:06.469><c>the </c><00:14:06.583><c>variability </c><00:14:06.697><c>of </c><00:14:06.811><c>the </c><00:14:06.925><c>actual </c><00:14:07.039><c>value.</c> 00:14:07.629 --> 00:14:10.590 align:start position:0% the identification of the model's overall trend and verification of how well the model tracks the variability of the actual value. 00:14:10.590 --> 00:14:10.600 align:start position:0% 00:14:10.600 --> 00:14:13.629 align:start position:0% Based <00:14:10.875><c>on </c><00:14:11.150><c>this </c><00:14:11.425><c>graph, </c><00:14:11.700><c>it </c><00:14:11.975><c>can </c><00:14:12.250><c>be </c><00:14:12.525><c>concluded </c><00:14:12.800><c>that</c> 00:14:13.629 --> 00:14:16.310 align:start position:0% Based on this graph, it can be concluded that 00:14:16.310 --> 00:14:19.069 align:start position:0% 00:14:19.069 --> 00:14:19.079 align:start position:0% 00:14:19.079 --> 00:14:21.550 align:start position:0% training <00:14:19.232><c>the </c><00:14:19.385><c>dataset </c><00:14:19.538><c>using </c><00:14:19.691><c>an </c><00:14:19.844><c>ensemble </c><00:14:19.997><c>model </c><00:14:20.150><c>is </c><00:14:20.303><c>better </c><00:14:20.456><c>than </c><00:14:20.609><c>coding </c><00:14:20.762><c>TFT </c><00:14:20.915><c>alone. </c><00:14:21.068><c>This</c> 00:14:21.550 --> 00:14:21.560 align:start position:0% training the dataset using an ensemble model is better than coding TFT alone. This 00:14:21.560 --> 00:14:24.230 align:start position:0% training the dataset using an ensemble model is better than coding TFT alone. This graph <00:14:23.480><c>compares</c> 00:14:24.230 --> 00:14:24.240 align:start position:0% graph compares 00:14:24.240 --> 00:14:26.870 align:start position:0% graph compares the <00:14:24.472><c>predicted </c><00:14:24.704><c>values </c><00:14:24.936><c>of </c><00:14:25.168><c>five </c><00:14:25.400><c>randomly </c><00:14:25.632><c>selected </c><00:14:25.864><c>samples </c><00:14:26.096><c>with </c><00:14:26.328><c>the </c><00:14:26.560><c>actual</c> 00:14:26.870 --> 00:14:26.880 align:start position:0% the predicted values of five randomly selected samples with the actual 00:14:26.880 --> 00:14:29.430 align:start position:0% the predicted values of five randomly selected samples with the actual values. <00:14:27.133><c>It </c><00:14:27.386><c>allows </c><00:14:27.639><c>for </c><00:14:27.892><c>the </c><00:14:28.145><c>evaluation </c><00:14:28.398><c>of </c><00:14:28.651><c>the </c><00:14:28.904><c>model's </c><00:14:29.157><c>short-term</c> 00:14:29.430 --> 00:14:29.440 align:start position:0% values. It allows for the evaluation of the model's short-term 00:14:29.440 --> 00:14:31.870 align:start position:0% values. It allows for the evaluation of the model's short-term prediction <00:14:30.160><c>performance </c><00:14:30.880><c>and </c><00:14:31.600><c>shows</c> 00:14:31.870 --> 00:14:31.880 align:start position:0% prediction performance and shows 00:14:31.880 --> 00:14:34.749 align:start position:0% prediction performance and shows how <00:14:32.094><c>well </c><00:14:32.308><c>the </c><00:14:32.522><c>model </c><00:14:32.736><c>has </c><00:14:32.950><c>learned </c><00:14:33.164><c>specific </c><00:14:33.378><c>patterns </c><00:14:33.592><c>for </c><00:14:33.806><c>each </c><00:14:34.020><c>sample. </c><00:14:34.234><c>The</c> 00:14:34.749 --> 00:14:34.759 align:start position:0% how well the model has learned specific patterns for each sample. The 00:14:34.759 --> 00:14:35.749 align:start position:0% how well the model has learned specific patterns for each sample. The 00:14:35.749 --> 00:14:35.759 align:start position:0% 00:14:35.759 --> 00:14:39.189 align:start position:0% LSTM <00:14:38.480><c>model</c> 00:14:39.189 --> 00:14:40.910 align:start position:0% LSTM model 00:14:40.910 --> 00:14:40.920 align:start position:0% 00:14:40.920 --> 00:14:43.870 align:start position:0% is <00:14:41.042><c>a </c><00:14:41.164><c>deep </c><00:14:41.286><c>learning </c><00:14:41.408><c>model </c><00:14:41.530><c>suitable </c><00:14:41.652><c>for </c><00:14:41.774><c>predicting </c><00:14:41.896><c>the </c><00:14:42.018><c>future </c><00:14:42.140><c>by </c><00:14:42.262><c>learning </c><00:14:42.384><c>from </c><00:14:42.506><c>time </c><00:14:42.628><c>series </c><00:14:42.750><c>data. </c><00:14:42.872><c>In</c> 00:14:43.870 --> 00:14:43.880 align:start position:0% is a deep learning model suitable for predicting the future by learning from time series data. In 00:14:43.880 --> 00:14:47.110 align:start position:0% is a deep learning model suitable for predicting the future by learning from time series data. In this <00:14:44.376><c>study, </c><00:14:44.872><c>we </c><00:14:45.368><c>utilized </c><00:14:45.864><c>LSTM </c><00:14:46.360><c>to</c> 00:14:47.110 --> 00:14:49.590 align:start position:0% this study, we utilized LSTM to 00:14:49.590 --> 00:14:49.600 align:start position:0% 00:14:49.600 --> 00:14:50.910 align:start position:0% 00:14:50.910 --> 00:14:50.920 align:start position:0% 00:14:50.920 --> 00:14:53.550 align:start position:0% predict <00:14:51.073><c>future </c><00:14:51.226><c>inventory </c><00:14:51.379><c>levels </c><00:14:51.532><c>based </c><00:14:51.685><c>on </c><00:14:51.838><c>corporate </c><00:14:51.991><c>inventory </c><00:14:52.144><c>data. </c><00:14:52.297><c>This </c><00:14:52.450><c>graph </c><00:14:52.603><c>compares </c><00:14:52.756><c>the</c> 00:14:53.550 --> 00:14:53.560 align:start position:0% predict future inventory levels based on corporate inventory data. This graph compares the 00:14:53.560 --> 00:14:56.350 align:start position:0% predict future inventory levels based on corporate inventory data. This graph compares the inventory <00:14:53.684><c>levels </c><00:14:53.808><c>predicted </c><00:14:53.932><c>by </c><00:14:54.056><c>the </c><00:14:54.180><c>LSTM </c><00:14:54.304><c>model </c><00:14:54.428><c>with </c><00:14:54.552><c>the </c><00:14:54.676><c>actual </c><00:14:54.800><c>inventory </c><00:14:54.924><c>levels. </c><00:14:55.048><c>It </c><00:14:55.172><c>allows </c><00:14:55.296><c>for </c><00:14:55.420><c>an </c><00:14:55.544><c>intuitive </c><00:14:55.668><c>verification </c><00:14:55.792><c>of</c> 00:14:56.350 --> 00:14:59.030 align:start position:0% inventory levels predicted by the LSTM model with the actual inventory levels. It allows for an intuitive verification of 00:14:59.030 --> 00:14:59.040 align:start position:0% 00:14:59.040 --> 00:15:01.389 align:start position:0% how <00:14:59.243><c>well </c><00:14:59.446><c>the </c><00:14:59.649><c>LSTM </c><00:14:59.852><c>model </c><00:15:00.055><c>predicts </c><00:15:00.258><c>inventory </c><00:15:00.461><c>variability </c><00:15:00.664><c>over </c><00:15:00.867><c>time. </c><00:15:01.070><c>This</c> 00:15:01.389 --> 00:15:04.030 align:start position:0% how well the LSTM model predicts inventory variability over time. This 00:15:04.030 --> 00:15:04.040 align:start position:0% 00:15:04.040 --> 00:15:06.749 align:start position:0% graph <00:15:04.487><c>visually </c><00:15:04.934><c>demonstrates </c><00:15:05.381><c>the </c><00:15:05.828><c>performance </c><00:15:06.275><c>by</c> 00:15:06.749 --> 00:15:06.759 align:start position:0% graph visually demonstrates the performance by 00:15:06.759 --> 00:15:09.749 align:start position:0% graph visually demonstrates the performance by comparing <00:15:06.945><c>the </c><00:15:07.131><c>predicted </c><00:15:07.317><c>values </c><00:15:07.503><c>of </c><00:15:07.689><c>the </c><00:15:07.875><c>XGB </c><00:15:08.061><c>ensemble </c><00:15:08.247><c>model </c><00:15:08.433><c>with </c><00:15:08.619><c>the </c><00:15:08.805><c>actual </c><00:15:08.991><c>values.</c> 00:15:09.749 --> 00:15:13.629 align:start position:0% comparing the predicted values of the XGB ensemble model with the actual values. 00:15:13.629 --> 00:15:13.639 align:start position:0% 00:15:13.639 --> 00:15:16.509 align:start position:0% How <00:15:13.861><c>the </c><00:15:14.083><c>predicted </c><00:15:14.305><c>values </c><00:15:14.527><c>differ </c><00:15:14.749><c>from </c><00:15:14.971><c>the </c><00:15:15.193><c>actual </c><00:15:15.415><c>values </c><00:15:15.637><c>You</c> 00:15:16.509 --> 00:15:16.519 align:start position:0% How the predicted values differ from the actual values You 00:15:16.519 --> 00:15:19.110 align:start position:0% How the predicted values differ from the actual values You can <00:15:16.668><c>evaluate </c><00:15:16.817><c>the </c><00:15:16.966><c>accuracy </c><00:15:17.115><c>of </c><00:15:17.264><c>the </c><00:15:17.413><c>model </c><00:15:17.562><c>by </c><00:15:17.711><c>checking </c><00:15:17.860><c>for </c><00:15:18.009><c>similarity. </c><00:15:18.158><c>The</c> 00:15:19.110 --> 00:15:19.120 align:start position:0% can evaluate the accuracy of the model by checking for similarity. The 00:15:19.120 --> 00:15:22.069 align:start position:0% can evaluate the accuracy of the model by checking for similarity. The more <00:15:19.335><c>similar </c><00:15:19.550><c>the </c><00:15:19.765><c>predicted </c><00:15:19.980><c>value </c><00:15:20.195><c>is </c><00:15:20.410><c>to </c><00:15:20.625><c>the </c><00:15:20.840><c>actual </c><00:15:21.055><c>value, </c><00:15:21.270><c>the</c> 00:15:22.069 --> 00:15:22.079 align:start position:0% more similar the predicted value is to the actual value, the 00:15:22.079 --> 00:15:25.069 align:start position:0% more similar the predicted value is to the actual value, the better <00:15:22.567><c>the </c><00:15:23.055><c>model </c><00:15:23.543><c>performance. </c><00:15:24.031><c>This </c><00:15:24.519><c>graph</c> 00:15:25.069 --> 00:15:27.790 align:start position:0% better the model performance. This graph 00:15:27.790 --> 00:15:27.800 align:start position:0% 00:15:27.800 --> 00:15:31.269 align:start position:0% visually <00:15:27.996><c>displays </c><00:15:28.192><c>the </c><00:15:28.388><c>performance </c><00:15:28.584><c>of </c><00:15:28.780><c>the </c><00:15:28.976><c>XGB </c><00:15:29.172><c>modal </c><00:15:29.368><c>BL </c><00:15:29.564><c>model </c><00:15:29.760><c>by </c><00:15:29.956><c>comparing </c><00:15:30.152><c>its</c> 00:15:31.269 --> 00:15:31.279 align:start position:0% visually displays the performance of the XGB modal BL model by comparing its 00:15:31.279 --> 00:15:34.590 align:start position:0% visually displays the performance of the XGB modal BL model by comparing its predicted <00:15:31.603><c>values </c><00:15:31.927><c>with </c><00:15:32.251><c>the </c><00:15:32.575><c>actual </c><00:15:32.899><c>values. </c><00:15:33.223><c>You </c><00:15:33.547><c>can </c><00:15:33.871><c>evaluate </c><00:15:34.195><c>the</c> 00:15:34.590 --> 00:15:34.600 align:start position:0% predicted values with the actual values. You can evaluate the 00:15:34.600 --> 00:15:37.069 align:start position:0% predicted values with the actual values. You can evaluate the accuracy <00:15:34.715><c>of </c><00:15:34.830><c>the </c><00:15:34.945><c>model </c><00:15:35.060><c>by </c><00:15:35.175><c>checking </c><00:15:35.290><c>how </c><00:15:35.405><c>similar </c><00:15:35.520><c>the </c><00:15:35.635><c>predicted </c><00:15:35.750><c>value </c><00:15:35.865><c>is </c><00:15:35.980><c>to </c><00:15:36.095><c>the </c><00:15:36.210><c>actual </c><00:15:36.325><c>value. </c><00:15:36.440><c>The</c> 00:15:37.069 --> 00:15:39.790 align:start position:0% accuracy of the model by checking how similar the predicted value is to the actual value. The 00:15:39.790 --> 00:15:39.800 align:start position:0% 00:15:39.800 --> 00:15:43.030 align:start position:0% more <00:15:39.949><c>similar </c><00:15:40.098><c>the </c><00:15:40.247><c>predicted </c><00:15:40.396><c>value </c><00:15:40.545><c>is </c><00:15:40.694><c>to </c><00:15:40.843><c>the </c><00:15:40.992><c>actual </c><00:15:41.141><c>value, </c><00:15:41.290><c>the </c><00:15:41.439><c>better </c><00:15:41.588><c>the </c><00:15:41.737><c>model </c><00:15:41.886><c>performance. </c><00:15:42.035><c>This</c> 00:15:43.030 --> 00:15:43.040 align:start position:0% more similar the predicted value is to the actual value, the better the model performance. This 00:15:43.040 --> 00:15:45.870 align:start position:0% more similar the predicted value is to the actual value, the better the model performance. This graph 00:15:45.870 --> 00:15:45.880 align:start position:0% graph 00:15:45.880 --> 00:15:48.350 align:start position:0% graph visualizes <00:15:46.120><c>feature </c><00:15:46.360><c>importance </c><00:15:46.600><c>based </c><00:15:46.840><c>on </c><00:15:47.080><c>the </c><00:15:47.320><c>XGB </c><00:15:47.560><c>model </c><00:15:47.800><c>and</c> 00:15:48.350 --> 00:15:51.670 align:start position:0% visualizes feature importance based on the XGB model and 00:15:51.670 --> 00:15:51.680 align:start position:0% 00:15:51.680 --> 00:15:55.030 align:start position:0% shows <00:15:51.899><c>the </c><00:15:52.118><c>impact </c><00:15:52.337><c>each </c><00:15:52.556><c>feature </c><00:15:52.775><c>has </c><00:15:52.994><c>on </c><00:15:53.213><c>the </c><00:15:53.432><c>model's </c><00:15:53.651><c>prediction. </c><00:15:53.870><c>For </c><00:15:54.089><c>example, </c><00:15:54.308><c>the</c> 00:15:55.030 --> 00:15:55.040 align:start position:0% shows the impact each feature has on the model's prediction. For example, the 00:15:55.040 --> 00:15:57.470 align:start position:0% shows the impact each feature has on the model's prediction. For example, the characteristics <00:15:55.189><c>of </c><00:15:55.338><c>cod </c><00:15:55.487><c>and </c><00:15:55.636><c>price </c><00:15:55.785><c>were </c><00:15:55.934><c>found </c><00:15:56.083><c>to </c><00:15:56.232><c>have </c><00:15:56.381><c>the </c><00:15:56.530><c>greatest </c><00:15:56.679><c>impact </c><00:15:56.828><c>on </c><00:15:56.977><c>the </c><00:15:57.126><c>model's </c><00:15:57.275><c>prediction.</c> 00:15:57.470 --> 00:15:57.480 align:start position:0% characteristics of cod and price were found to have the greatest impact on the model's prediction. 00:15:57.480 --> 00:16:00.189 align:start position:0% characteristics of cod and price were found to have the greatest impact on the model's prediction. This <00:15:58.053><c>allows </c><00:15:58.626><c>you </c><00:15:59.199><c>to</c> 00:16:00.189 --> 00:16:02.710 align:start position:0% This allows you to 00:16:02.710 --> 00:16:02.720 align:start position:0% 00:16:02.720 --> 00:16:04.990 align:start position:0% visually <00:16:02.959><c>confirm </c><00:16:03.198><c>which </c><00:16:03.437><c>characteristics </c><00:16:03.676><c>the </c><00:16:03.915><c>model </c><00:16:04.154><c>considers </c><00:16:04.393><c>important. </c><00:16:04.632><c>This</c> 00:16:04.990 --> 00:16:05.000 align:start position:0% visually confirm which characteristics the model considers important. This 00:16:05.000 --> 00:16:07.949 align:start position:0% visually confirm which characteristics the model considers important. This graph <00:16:05.279><c>visualizes </c><00:16:05.558><c>the </c><00:16:05.837><c>feature </c><00:16:06.116><c>importance </c><00:16:06.395><c>of </c><00:16:06.674><c>the </c><00:16:06.953><c>ensemble </c><00:16:07.232><c>model. </c><00:16:07.511><c>The</c> 00:16:07.949 --> 00:16:10.710 align:start position:0% graph visualizes the feature importance of the ensemble model. The 00:16:10.710 --> 00:16:10.720 align:start position:0% 00:16:10.720 --> 00:16:13.829 align:start position:0% horizontal <00:16:11.143><c>axis </c><00:16:11.566><c>represents </c><00:16:11.989><c>feature </c><00:16:12.412><c>importance, </c><00:16:12.835><c>and</c> 00:16:13.829 --> 00:16:13.839 align:start position:0% horizontal axis represents feature importance, and 00:16:13.839 --> 00:16:17.069 align:start position:0% horizontal axis represents feature importance, and the <00:16:14.053><c>longer </c><00:16:14.267><c>the </c><00:16:14.481><c>bar, </c><00:16:14.695><c>the </c><00:16:14.909><c>greater </c><00:16:15.123><c>the </c><00:16:15.337><c>impact </c><00:16:15.551><c>the </c><00:16:15.765><c>feature </c><00:16:15.979><c>has </c><00:16:16.193><c>on </c><00:16:16.407><c>the </c><00:16:16.621><c>model's </c><00:16:16.835><c>prediction.</c> 00:16:17.069 --> 00:16:17.079 align:start position:0% the longer the bar, the greater the impact the feature has on the model's prediction. 00:16:17.079 --> 00:16:20.230 align:start position:0% the longer the bar, the greater the impact the feature has on the model's prediction. Inventory 00:16:20.230 --> 00:16:20.240 align:start position:0% Inventory 00:16:20.240 --> 00:16:22.550 align:start position:0% Inventory was <00:16:20.548><c>found </c><00:16:20.856><c>to </c><00:16:21.164><c>be </c><00:16:21.472><c>the </c><00:16:21.780><c>most </c><00:16:22.088><c>important </c><00:16:22.396><c>characteristic,</c> 00:16:22.550 --> 00:16:22.560 align:start position:0% was found to be the most important characteristic, 00:16:22.560 --> 00:16:24.790 align:start position:0% was found to be the most important characteristic, followed <00:16:22.927><c>by </c><00:16:23.294><c>weather </c><00:16:23.661><c>and </c><00:16:24.028><c>price, </c><00:16:24.395><c>which</c> 00:16:24.790 --> 00:16:26.389 align:start position:0% followed by weather and price, which 00:16:26.389 --> 00:16:26.399 align:start position:0% 00:16:26.399 --> 00:16:29.150 align:start position:0% were <00:16:26.573><c>analyzed </c><00:16:26.747><c>to </c><00:16:26.921><c>have </c><00:16:27.095><c>a </c><00:16:27.269><c>significant </c><00:16:27.443><c>impact. </c><00:16:27.617><c>Through </c><00:16:27.791><c>this, </c><00:16:27.965><c>you </c><00:16:28.139><c>can </c><00:16:28.313><c>see</c> 00:16:29.150 --> 00:16:29.160 align:start position:0% were analyzed to have a significant impact. Through this, you can see 00:16:29.160 --> 00:16:31.030 align:start position:0% were analyzed to have a significant impact. Through this, you can see which <00:16:29.300><c>characteristics </c><00:16:29.440><c>the </c><00:16:29.580><c>ensemble </c><00:16:29.720><c>model </c><00:16:29.860><c>focuses </c><00:16:30.000><c>on </c><00:16:30.140><c>to </c><00:16:30.280><c>perform </c><00:16:30.420><c>predictions. </c><00:16:30.560><c>This</c> 00:16:31.030 --> 00:16:33.509 align:start position:0% which characteristics the ensemble model focuses on to perform predictions. This 00:16:33.509 --> 00:16:33.519 align:start position:0% 00:16:33.519 --> 00:16:36.629 align:start position:0% graph <00:16:33.713><c>shows </c><00:16:33.907><c>the </c><00:16:34.101><c>changes </c><00:16:34.295><c>in </c><00:16:34.489><c>training </c><00:16:34.683><c>and </c><00:16:34.877><c>validation </c><00:16:35.071><c>losses </c><00:16:35.265><c>of </c><00:16:35.459><c>the </c><00:16:35.653><c>ensemble </c><00:16:35.847><c>model. </c><00:16:36.041><c>Although </c><00:16:36.235><c>the</c> 00:16:36.629 --> 00:16:36.639 align:start position:0% graph shows the changes in training and validation losses of the ensemble model. Although the 00:16:36.639 --> 00:16:39.309 align:start position:0% graph shows the changes in training and validation losses of the ensemble model. Although the initial <00:16:38.959><c>loss</c> 00:16:39.309 --> 00:16:39.319 align:start position:0% initial loss 00:16:39.319 --> 00:16:42.150 align:start position:0% initial loss value <00:16:39.639><c>is </c><00:16:39.959><c>somewhat </c><00:16:40.279><c>high, </c><00:16:40.599><c>you </c><00:16:40.919><c>can </c><00:16:41.239><c>observe </c><00:16:41.559><c>a</c> 00:16:42.150 --> 00:16:42.160 align:start position:0% value is somewhat high, you can observe a 00:16:42.160 --> 00:16:44.829 align:start position:0% value is somewhat high, you can observe a trend <00:16:42.420><c>where </c><00:16:42.680><c>the </c><00:16:42.940><c>loss </c><00:16:43.200><c>steadily </c><00:16:43.460><c>decreases </c><00:16:43.720><c>as </c><00:16:43.980><c>training </c><00:16:44.240><c>progresses.</c> 00:16:44.829 --> 00:16:44.839 align:start position:0% trend where the loss steadily decreases as training progresses. 00:16:44.839 --> 00:16:48.230 align:start position:0% trend where the loss steadily decreases as training progresses. In <00:16:45.319><c>particular, </c><00:16:45.799><c>the </c><00:16:46.279><c>training </c><00:16:46.759><c>loss </c><00:16:47.239><c>and</c> 00:16:48.230 --> 00:16:48.240 align:start position:0% In particular, the training loss and 00:16:48.240 --> 00:16:50.670 align:start position:0% In particular, the training loss and validation <00:16:48.400><c>loss </c><00:16:48.560><c>are </c><00:16:48.720><c>similar. </c><00:16:48.880><c>As </c><00:16:49.040><c>it </c><00:16:49.200><c>is </c><00:16:49.360><c>reduced </c><00:16:49.520><c>into </c><00:16:49.680><c>a </c><00:16:49.840><c>pattern, </c><00:16:50.000><c>it</c> 00:16:50.670 --> 00:16:53.030 align:start position:0% validation loss are similar. As it is reduced into a pattern, it 00:16:53.030 --> 00:16:53.040 align:start position:0% 00:16:53.040 --> 00:16:55.590 align:start position:0% indicates <00:16:53.213><c>that </c><00:16:53.386><c>the </c><00:16:53.559><c>model </c><00:16:53.732><c>is </c><00:16:53.905><c>stably </c><00:16:54.078><c>learning </c><00:16:54.251><c>the </c><00:16:54.424><c>data </c><00:16:54.597><c>patterns.</c> 00:16:55.590 --> 00:16:55.600 align:start position:0% indicates that the model is stably learning the data patterns. 00:16:55.600 --> 00:16:58.870 align:start position:0% indicates that the model is stably learning the data patterns. Through <00:16:55.757><c>this, </c><00:16:55.914><c>it </c><00:16:56.071><c>can </c><00:16:56.228><c>be </c><00:16:56.385><c>concluded </c><00:16:56.542><c>that </c><00:16:56.699><c>the </c><00:16:56.856><c>ensemble </c><00:16:57.013><c>model </c><00:16:57.170><c>is </c><00:16:57.327><c>learning </c><00:16:57.484><c>while </c><00:16:57.641><c>maintaining </c><00:16:57.798><c>general </c><00:16:57.955><c>performance </c><00:16:58.112><c>without </c><00:16:58.269><c>overfitting. </c><00:16:58.426><c>This</c> 00:16:58.870 --> 00:17:01.389 align:start position:0% Through this, it can be concluded that the ensemble model is learning while maintaining general performance without overfitting. This 00:17:01.389 --> 00:17:01.399 align:start position:0% 00:17:01.399 --> 00:17:03.790 align:start position:0% graph 00:17:03.790 --> 00:17:06.110 align:start position:0% graph 00:17:06.110 --> 00:17:06.120 align:start position:0% 00:17:06.120 --> 00:17:08.350 align:start position:0% visualizes <00:17:06.253><c>demand </c><00:17:06.386><c>fluctuations </c><00:17:06.519><c>based </c><00:17:06.652><c>on </c><00:17:06.785><c>holidays </c><00:17:06.918><c>and </c><00:17:07.051><c>price </c><00:17:07.184><c>changes </c><00:17:07.317><c>using </c><00:17:07.450><c>a </c><00:17:07.583><c>decoder. </c><00:17:07.716><c>It</c> 00:17:08.350 --> 00:17:11.750 align:start position:0% visualizes demand fluctuations based on holidays and price changes using a decoder. It 00:17:11.750 --> 00:17:14.870 align:start position:0% 00:17:14.870 --> 00:17:14.880 align:start position:0% 00:17:14.880 --> 00:17:17.549 align:start position:0% intuitively <00:17:14.969><c>shows </c><00:17:15.058><c>how </c><00:17:15.147><c>demand </c><00:17:15.236><c>varies </c><00:17:15.325><c>by </c><00:17:15.414><c>price </c><00:17:15.503><c>range </c><00:17:15.592><c>and </c><00:17:15.681><c>various </c><00:17:15.770><c>holidays </c><00:17:15.859><c>such </c><00:17:15.948><c>as </c><00:17:16.037><c>Children's </c><00:17:16.126><c>Day </c><00:17:16.215><c>and </c><00:17:16.304><c>Chuseok.</c> 00:17:17.549 --> 00:17:17.559 align:start position:0% intuitively shows how demand varies by price range and various holidays such as Children's Day and Chuseok. 00:17:17.559 --> 00:17:20.189 align:start position:0% intuitively shows how demand varies by price range and various holidays such as Children's Day and Chuseok. This 00:17:20.189 --> 00:17:20.199 align:start position:0% This 00:17:20.199 --> 00:17:23.390 align:start position:0% This provides <00:17:20.513><c>useful </c><00:17:20.827><c>data </c><00:17:21.141><c>for </c><00:17:21.455><c>establishing </c><00:17:21.769><c>marketing </c><00:17:22.083><c>strategies. </c><00:17:22.397><c>This</c> 00:17:23.390 --> 00:17:23.400 align:start position:0% provides useful data for establishing marketing strategies. This 00:17:23.400 --> 00:17:25.510 align:start position:0% provides useful data for establishing marketing strategies. This graph <00:17:23.890><c>evaluates </c><00:17:24.380><c>model </c><00:17:24.870><c>performance </c><00:17:25.360><c>by</c> 00:17:25.510 --> 00:17:25.520 align:start position:0% graph evaluates model performance by 00:17:25.520 --> 00:17:28.429 align:start position:0% graph evaluates model performance by comparing <00:17:25.669><c>the </c><00:17:25.818><c>actual </c><00:17:25.967><c>and </c><00:17:26.116><c>predicted </c><00:17:26.265><c>values </c><00:17:26.414><c>of </c><00:17:26.563><c>five </c><00:17:26.712><c>randomly </c><00:17:26.861><c>selected </c><00:17:27.010><c>samples. </c><00:17:27.159><c>It </c><00:17:27.308><c>allows </c><00:17:27.457><c>you </c><00:17:27.606><c>to </c><00:17:27.755><c>verify</c> 00:17:28.429 --> 00:17:31.549 align:start position:0% comparing the actual and predicted values of five randomly selected samples. It allows you to verify 00:17:31.549 --> 00:17:31.559 align:start position:0% 00:17:31.559 --> 00:17:33.789 align:start position:0% how <00:17:31.754><c>well </c><00:17:31.949><c>the </c><00:17:32.144><c>model </c><00:17:32.339><c>has </c><00:17:32.534><c>learned </c><00:17:32.729><c>specific </c><00:17:32.924><c>patterns. </c><00:17:33.119><c>This</c> 00:17:33.789 --> 00:17:33.799 align:start position:0% how well the model has learned specific patterns. This 00:17:33.799 --> 00:17:36.830 align:start position:0% how well the model has learned specific patterns. This graph <00:17:34.088><c>demonstrates </c><00:17:34.377><c>the </c><00:17:34.666><c>efficiency </c><00:17:34.955><c>of </c><00:17:35.244><c>inventory </c><00:17:35.533><c>management </c><00:17:35.822><c>by </c><00:17:36.111><c>visualizing </c><00:17:36.400><c>optimal</c> 00:17:36.830 --> 00:17:36.840 align:start position:0% graph demonstrates the efficiency of inventory management by visualizing optimal 00:17:36.840 --> 00:17:39.549 align:start position:0% graph demonstrates the efficiency of inventory management by visualizing optimal inventory <00:17:37.269><c>levels </c><00:17:37.698><c>and </c><00:17:38.127><c>predicted </c><00:17:38.556><c>demand.</c> 00:17:39.549 --> 00:17:42.350 align:start position:0% inventory levels and predicted demand. 00:17:42.350 --> 00:17:42.360 align:start position:0% 00:17:42.360 --> 00:17:45.070 align:start position:0% Considering <00:17:42.532><c>safety </c><00:17:42.704><c>stock </c><00:17:42.876><c>levels, </c><00:17:43.048><c>an </c><00:17:43.220><c>appropriate </c><00:17:43.392><c>inventory </c><00:17:43.564><c>level </c><00:17:43.736><c>capable </c><00:17:43.908><c>of </c><00:17:44.080><c>responding </c><00:17:44.252><c>to </c><00:17:44.424><c>demand </c><00:17:44.596><c>fluctuations</c> 00:17:45.070 --> 00:17:46.789 align:start position:0% Considering safety stock levels, an appropriate inventory level capable of responding to demand fluctuations 00:17:46.789 --> 00:17:46.799 align:start position:0% 00:17:46.799 --> 00:17:49.710 align:start position:0% was <00:17:47.073><c>set, </c><00:17:47.347><c>and </c><00:17:47.621><c>the </c><00:17:47.895><c>estimated </c><00:17:48.169><c>safety </c><00:17:48.443><c>stock </c><00:17:48.717><c>level</c> 00:17:49.710 --> 00:17:49.720 align:start position:0% was set, and the estimated safety stock level 00:17:49.720 --> 00:17:53.190 align:start position:0% was set, and the estimated safety stock level is <00:17:50.060><c>34.02 </c><00:17:50.400><c>units. </c><00:17:50.740><c>The </c><00:17:51.080><c>analysis </c><00:17:51.420><c>results </c><00:17:51.760><c>showed </c><00:17:52.100><c>that </c><00:17:52.440><c>the</c> 00:17:53.190 --> 00:17:53.200 align:start position:0% is 34.02 units. The analysis results showed that the 00:17:53.200 --> 00:17:56.470 align:start position:0% is 34.02 units. The analysis results showed that the TFT <00:17:53.640><c>model </c><00:17:54.080><c>and </c><00:17:54.520><c>the </c><00:17:54.960><c>xgb </c><00:17:55.400><c>positive </c><00:17:55.840><c>model</c> 00:17:56.470 --> 00:17:56.480 align:start position:0% TFT model and the xgb positive model 00:17:56.480 --> 00:17:58.830 align:start position:0% TFT model and the xgb positive model demonstrated <00:17:56.786><c>excellent </c><00:17:57.092><c>performance; </c><00:17:57.398><c>in </c><00:17:57.704><c>particular, </c><00:17:58.010><c>the </c><00:17:58.316><c>xgb</c> 00:17:58.830 --> 00:17:58.840 align:start position:0% demonstrated excellent performance; in particular, the xgb 00:17:58.840 --> 00:18:01.549 align:start position:0% demonstrated excellent performance; in particular, the xgb non-positive <00:17:59.919><c>model</c> 00:18:01.549 --> 00:18:03.950 align:start position:0% non-positive model 00:18:03.950 --> 00:18:03.960 align:start position:0% 00:18:03.960 --> 00:18:06.310 align:start position:0% recorded <00:18:04.071><c>the </c><00:18:04.182><c>lowest </c><00:18:04.293><c>error </c><00:18:04.404><c>rate </c><00:18:04.515><c>of </c><00:18:04.626><c>17.7%. </c><00:18:04.737><c>In </c><00:18:04.848><c>the </c><00:18:04.959><c>future, </c><00:18:05.070><c>we </c><00:18:05.181><c>plan </c><00:18:05.292><c>to </c><00:18:05.403><c>further </c><00:18:05.514><c>improve </c><00:18:05.625><c>model </c><00:18:05.736><c>performance </c><00:18:05.847><c>by </c><00:18:05.958><c>addressing </c><00:18:06.069><c>the</c> 00:18:06.310 --> 00:18:06.320 align:start position:0% recorded the lowest error rate of 17.7%. In the future, we plan to further improve model performance by addressing the 00:18:06.320 --> 00:18:09.230 align:start position:0% recorded the lowest error rate of 17.7%. In the future, we plan to further improve model performance by addressing the issue <00:18:06.545><c>of </c><00:18:06.770><c>resistance </c><00:18:06.995><c>to </c><00:18:07.220><c>learning </c><00:18:07.445><c>speed </c><00:18:07.670><c>due </c><00:18:07.895><c>to </c><00:18:08.120><c>increased </c><00:18:08.345><c>data </c><00:18:08.570><c>volume </c><00:18:08.795><c>and</c> 00:18:09.230 --> 00:18:09.240 align:start position:0% issue of resistance to learning speed due to increased data volume and 00:18:09.240 --> 00:18:11.830 align:start position:0% issue of resistance to learning speed due to increased data volume and strengthening <00:18:09.546><c>the </c><00:18:09.852><c>reflection </c><00:18:10.158><c>of </c><00:18:10.464><c>external </c><00:18:10.770><c>variables. </c><00:18:11.076><c>This</c> 00:18:11.830 --> 00:18:15.029 align:start position:0% strengthening the reflection of external variables. This 00:18:15.029 --> 00:18:15.039 align:start position:0% 00:18:15.039 --> 00:18:17.870 align:start position:0% demand <00:18:15.219><c>forecasting </c><00:18:15.399><c>model </c><00:18:15.579><c>can </c><00:18:15.759><c>be </c><00:18:15.939><c>effectively </c><00:18:16.119><c>utilized </c><00:18:16.299><c>for </c><00:18:16.479><c>corporate </c><00:18:16.659><c>inventory </c><00:18:16.839><c>management </c><00:18:17.019><c>and </c><00:18:17.199><c>the</c> 00:18:17.870 --> 00:18:17.880 align:start position:0% demand forecasting model can be effectively utilized for corporate inventory management and the 00:18:17.880 --> 00:18:20.789 align:start position:0% demand forecasting model can be effectively utilized for corporate inventory management and the establishment <00:18:18.480><c>of </c><00:18:19.080><c>marketing </c><00:18:19.680><c>strategies. </c><00:18:20.280><c>This</c> 00:18:20.789 --> 00:18:20.799 align:start position:0% establishment of marketing strategies. This 00:18:20.799 --> 00:18:24.190 align:start position:0% establishment of marketing strategies. This chart <00:18:21.407><c>shows </c><00:18:22.015><c>the </c><00:18:22.623><c>artificial </c><00:18:23.231><c>neural </c><00:18:23.839><c>network</c> 00:18:24.190 --> 00:18:24.200 align:start position:0% chart shows the artificial neural network 00:18:24.200 --> 00:18:26.830 align:start position:0% chart shows the artificial neural network model <00:18:24.338><c>learning </c><00:18:24.476><c>It </c><00:18:24.614><c>demonstrates </c><00:18:24.752><c>that </c><00:18:24.890><c>the </c><00:18:25.028><c>loss </c><00:18:25.166><c>value </c><00:18:25.304><c>was </c><00:18:25.442><c>stably </c><00:18:25.580><c>reduced </c><00:18:25.718><c>during </c><00:18:25.856><c>the </c><00:18:25.994><c>process,</c> 00:18:26.830 --> 00:18:29.909 align:start position:0% model learning It demonstrates that the loss value was stably reduced during the process, 00:18:29.909 --> 00:18:32.190 align:start position:0% 00:18:32.190 --> 00:18:34.909 align:start position:0% 00:18:34.909 --> 00:18:34.919 align:start position:0% 00:18:34.919 --> 00:18:37.830 align:start position:0% suggesting <00:18:34.994><c>that </c><00:18:35.069><c>the </c><00:18:35.144><c>model </c><00:18:35.219><c>can </c><00:18:35.294><c>produce </c><00:18:35.369><c>good </c><00:18:35.444><c>predictive </c><00:18:35.519><c>performance </c><00:18:35.594><c>on </c><00:18:35.669><c>the </c><00:18:35.744><c>given </c><00:18:35.819><c>data. </c><00:18:35.894><c>This </c><00:18:35.969><c>is </c><00:18:36.044><c>a </c><00:18:36.119><c>chart </c><00:18:36.194><c>visualizing </c><00:18:36.269><c>the </c><00:18:36.344><c>predicted </c><00:18:36.419><c>demand </c><00:18:36.494><c>using </c><00:18:36.569><c>an </c><00:18:36.644><c>artificial </c><00:18:36.719><c>neural </c><00:18:36.794><c>network, </c><00:18:36.869><c>the</c> 00:18:37.830 --> 00:18:37.840 align:start position:0% suggesting that the model can produce good predictive performance on the given data. This is a chart visualizing the predicted demand using an artificial neural network, the 00:18:37.840 --> 00:18:41.070 align:start position:0% suggesting that the model can produce good predictive performance on the given data. This is a chart visualizing the predicted demand using an artificial neural network, the safety <00:18:38.706><c>stock, </c><00:18:39.572><c>and </c><00:18:40.438><c>the</c> 00:18:41.070 --> 00:18:41.080 align:start position:0% safety stock, and the 00:18:41.080 --> 00:18:43.470 align:start position:0% safety stock, and the optimal <00:18:41.331><c>stock </c><00:18:41.582><c>level </c><00:18:41.833><c>calculated </c><00:18:42.084><c>by </c><00:18:42.335><c>summing </c><00:18:42.586><c>them. </c><00:18:42.837><c>The</c> 00:18:43.470 --> 00:18:43.480 align:start position:0% optimal stock level calculated by summing them. The 00:18:43.480 --> 00:18:46.230 align:start position:0% optimal stock level calculated by summing them. The optimal <00:18:44.699><c>stock </c><00:18:45.918><c>level</c> 00:18:46.230 --> 00:18:46.240 align:start position:0% optimal stock level 00:18:46.240 --> 00:18:48.710 align:start position:0% optimal stock level varies <00:18:46.425><c>significantly </c><00:18:46.610><c>depending </c><00:18:46.795><c>on </c><00:18:46.980><c>fluctuations </c><00:18:47.165><c>in </c><00:18:47.350><c>predicted </c><00:18:47.535><c>demand; </c><00:18:47.720><c>while</c> 00:18:48.710 --> 00:18:48.720 align:start position:0% varies significantly depending on fluctuations in predicted demand; while 00:18:48.720 --> 00:18:51.029 align:start position:0% varies significantly depending on fluctuations in predicted demand; while predicted <00:18:50.640><c>demand</c> 00:18:51.029 --> 00:18:51.039 align:start position:0% predicted demand 00:18:51.039 --> 00:18:53.870 align:start position:0% predicted demand shows <00:18:51.235><c>a </c><00:18:51.431><c>relatively </c><00:18:51.627><c>constant </c><00:18:51.823><c>pattern, </c><00:18:52.019><c>it </c><00:18:52.215><c>exhibits </c><00:18:52.411><c>large </c><00:18:52.607><c>fluctuations </c><00:18:52.803><c>in </c><00:18:52.999><c>some </c><00:18:53.195><c>sections. </c><00:18:53.391><c>The</c> 00:18:53.870 --> 00:18:53.880 align:start position:0% shows a relatively constant pattern, it exhibits large fluctuations in some sections. The 00:18:53.880 --> 00:18:56.510 align:start position:0% shows a relatively constant pattern, it exhibits large fluctuations in some sections. The safety <00:18:55.919><c>stock</c> 00:18:56.510 --> 00:18:56.520 align:start position:0% safety stock 00:18:56.520 --> 00:18:59.149 align:start position:0% safety stock level <00:18:56.893><c>remains </c><00:18:57.266><c>stable, </c><00:18:57.639><c>allowing </c><00:18:58.012><c>the </c><00:18:58.385><c>company </c><00:18:58.758><c>to</c> 00:18:59.149 --> 00:19:02.070 align:start position:0% level remains stable, allowing the company to 00:19:02.070 --> 00:19:02.080 align:start position:0% 00:19:02.080 --> 00:19:05.270 align:start position:0% prepare <00:19:02.405><c>for </c><00:19:02.730><c>sudden </c><00:19:03.055><c>demand </c><00:19:03.380><c>fluctuations </c><00:19:03.705><c>or </c><00:19:04.030><c>supply </c><00:19:04.355><c>disruptions.</c> 00:19:05.270 --> 00:19:05.280 align:start position:0% prepare for sudden demand fluctuations or supply disruptions. 00:19:05.280 --> 00:19:08.310 align:start position:0% prepare for sudden demand fluctuations or supply disruptions. Visualization <00:19:05.564><c>of </c><00:19:05.848><c>training </c><00:19:06.132><c>history </c><00:19:06.416><c>using </c><00:19:06.700><c>SVM </c><00:19:06.984><c>demonstrates </c><00:19:07.268><c>high </c><00:19:07.552><c>prediction </c><00:19:07.836><c>accuracy</c> 00:19:08.310 --> 00:19:08.320 align:start position:0% Visualization of training history using SVM demonstrates high prediction accuracy 00:19:08.320 --> 00:19:10.590 align:start position:0% Visualization of training history using SVM demonstrates high prediction accuracy and <00:19:08.508><c>generalization. </c><00:19:08.696><c>The </c><00:19:08.884><c>actual </c><00:19:09.072><c>and </c><00:19:09.260><c>predicted </c><00:19:09.448><c>values </c><00:19:09.636><c>are</c> 00:19:10.590 --> 00:19:12.830 align:start position:0% and generalization. The actual and predicted values are 00:19:12.830 --> 00:19:12.840 align:start position:0% 00:19:12.840 --> 00:19:15.789 align:start position:0% distributed <00:19:13.018><c>close </c><00:19:13.196><c>to </c><00:19:13.374><c>the </c><00:19:13.552><c>x-line, </c><00:19:13.730><c>demonstrating </c><00:19:13.908><c>the </c><00:19:14.086><c>model's </c><00:19:14.264><c>high </c><00:19:14.442><c>predictive </c><00:19:14.620><c>performance </c><00:19:14.798><c>and</c> 00:19:15.789 --> 00:19:15.799 align:start position:0% distributed close to the x-line, demonstrating the model's high predictive performance and 00:19:15.799 --> 00:19:18.630 align:start position:0% distributed close to the x-line, demonstrating the model's high predictive performance and excellent <00:19:16.719><c>generalization </c><00:19:17.639><c>ability.</c> 00:19:18.630 --> 00:19:18.640 align:start position:0% excellent generalization ability. 00:19:18.640 --> 00:19:21.510 align:start position:0% excellent generalization ability. No <00:19:18.836><c>overfitting: </c><00:19:19.032><c>As </c><00:19:19.228><c>the </c><00:19:19.424><c>predictive </c><00:19:19.620><c>performance </c><00:19:19.816><c>of </c><00:19:20.012><c>the </c><00:19:20.208><c>training </c><00:19:20.404><c>and </c><00:19:20.600><c>test </c><00:19:20.796><c>data </c><00:19:20.992><c>is</c> 00:19:21.510 --> 00:19:21.520 align:start position:0% No overfitting: As the predictive performance of the training and test data is 00:19:21.520 --> 00:19:24.390 align:start position:0% No overfitting: As the predictive performance of the training and test data is similar, <00:19:21.927><c>no </c><00:19:22.334><c>signs </c><00:19:22.741><c>of </c><00:19:23.148><c>overfitting </c><00:19:23.555><c>are</c> 00:19:24.390 --> 00:19:24.400 align:start position:0% similar, no signs of overfitting are 00:19:24.400 --> 00:19:27.350 align:start position:0% similar, no signs of overfitting are present. <00:19:25.059><c>Presence </c><00:19:25.718><c>of </c><00:19:26.377><c>spectra </c><00:19:27.036><c>and</c> 00:19:27.350 --> 00:19:27.360 align:start position:0% present. Presence of spectra and 00:19:27.360 --> 00:19:30.549 align:start position:0% present. Presence of spectra and potential <00:19:27.517><c>for </c><00:19:27.674><c>model </c><00:19:27.831><c>improvement: </c><00:19:27.988><c>Predictive </c><00:19:28.145><c>spectra </c><00:19:28.302><c>exist, </c><00:19:28.459><c>and </c><00:19:28.616><c>model </c><00:19:28.773><c>performance </c><00:19:28.930><c>can </c><00:19:29.087><c>be </c><00:19:29.244><c>further </c><00:19:29.401><c>improved </c><00:19:29.558><c>through</c> 00:19:30.549 --> 00:19:30.559 align:start position:0% potential for model improvement: Predictive spectra exist, and model performance can be further improved through 00:19:30.559 --> 00:19:33.110 align:start position:0% potential for model improvement: Predictive spectra exist, and model performance can be further improved through spectra <00:19:31.059><c>distribution </c><00:19:31.559><c>analysis </c><00:19:32.059><c>and </c><00:19:32.559><c>hyperparameter</c> 00:19:33.110 --> 00:19:33.120 align:start position:0% spectra distribution analysis and hyperparameter 00:19:33.120 --> 00:19:35.630 align:start position:0% spectra distribution analysis and hyperparameter tuning. 00:19:35.630 --> 00:19:35.640 align:start position:0% tuning. 00:19:35.640 --> 00:19:38.029 align:start position:0% tuning. All <00:19:36.179><c>new </c><00:19:36.718><c>models </c><00:19:37.257><c>show </c><00:19:37.796><c>a</c> 00:19:38.029 --> 00:19:38.039 align:start position:0% All new models show a 00:19:38.039 --> 00:19:40.029 align:start position:0% All new models show a similarly <00:19:38.351><c>relatively </c><00:19:38.663><c>constant </c><00:19:38.975><c>pattern, </c><00:19:39.287><c>but </c><00:19:39.599><c>exhibit</c> 00:19:40.029 --> 00:19:40.039 align:start position:0% similarly relatively constant pattern, but exhibit 00:19:40.039 --> 00:19:42.710 align:start position:0% similarly relatively constant pattern, but exhibit large <00:19:40.487><c>fluctuations </c><00:19:40.935><c>in </c><00:19:41.383><c>some </c><00:19:41.831><c>sections. </c><00:19:42.279><c>The</c> 00:19:42.710 --> 00:19:42.720 align:start position:0% large fluctuations in some sections. The 00:19:42.720 --> 00:19:45.549 align:start position:0% large fluctuations in some sections. The XGB <00:19:43.329><c>model </c><00:19:43.938><c>demonstrates </c><00:19:44.547><c>excellent </c><00:19:45.156><c>predictive</c> 00:19:45.549 --> 00:19:45.559 align:start position:0% XGB model demonstrates excellent predictive 00:19:45.559 --> 00:19:47.789 align:start position:0% XGB model demonstrates excellent predictive performance <00:19:45.887><c>and </c><00:19:46.215><c>fast </c><00:19:46.543><c>training </c><00:19:46.871><c>speed </c><00:19:47.199><c>and</c> 00:19:47.789 --> 00:19:47.799 align:start position:0% performance and fast training speed and 00:19:47.799 --> 00:19:49.950 align:start position:0% performance and fast training speed and can <00:19:48.079><c>be </c><00:19:48.359><c>applied </c><00:19:48.639><c>to </c><00:19:48.919><c>various </c><00:19:49.199><c>types </c><00:19:49.479><c>of </c><00:19:49.759><c>data.</c> 00:19:49.950 --> 00:19:49.960 align:start position:0% can be applied to various types of data. 00:19:49.960 --> 00:19:53.350 align:start position:0% can be applied to various types of data. However, <00:19:50.314><c>there </c><00:19:50.668><c>is </c><00:19:51.022><c>a </c><00:19:51.376><c>risk </c><00:19:51.730><c>of </c><00:19:52.084><c>overfitting, </c><00:19:52.438><c>and</c> 00:19:53.350 --> 00:19:53.360 align:start position:0% However, there is a risk of overfitting, and 00:19:53.360 --> 00:19:55.909 align:start position:0% However, there is a risk of overfitting, and hyperparameter <00:19:53.879><c>tuning </c><00:19:54.398><c>is </c><00:19:54.917><c>required.</c> 00:19:55.909 --> 00:19:55.919 align:start position:0% hyperparameter tuning is required. 00:19:55.919 --> 00:19:59.230 align:start position:0% hyperparameter tuning is required. The <00:19:56.163><c>model </c><00:19:56.407><c>leverages </c><00:19:56.651><c>its </c><00:19:56.895><c>strength </c><00:19:57.139><c>in </c><00:19:57.383><c>learning </c><00:19:57.627><c>complex </c><00:19:57.871><c>non-linear </c><00:19:58.115><c>relationships.</c> 00:19:59.230 --> 00:19:59.240 align:start position:0% The model leverages its strength in learning complex non-linear relationships. 00:19:59.240 --> 00:20:02.070 align:start position:0% The model leverages its strength in learning complex non-linear relationships. Although <00:20:00.160><c>visible, </c><00:20:01.080><c>hyperparameter</c> 00:20:02.070 --> 00:20:02.080 align:start position:0% Although visible, hyperparameter 00:20:02.080 --> 00:20:04.750 align:start position:0% Although visible, hyperparameter tuning <00:20:02.826><c>is </c><00:20:03.572><c>crucial </c><00:20:04.318><c>and</c> 00:20:04.750 --> 00:20:04.760 align:start position:0% tuning is crucial and 00:20:04.760 --> 00:20:07.510 align:start position:0% tuning is crucial and there <00:20:05.114><c>is </c><00:20:05.468><c>a </c><00:20:05.822><c>risk </c><00:20:06.176><c>of </c><00:20:06.530><c>overfitting. </c><00:20:06.884><c>SVM </c><00:20:07.238><c>models</c> 00:20:07.510 --> 00:20:07.520 align:start position:0% there is a risk of overfitting. SVM models 00:20:07.520 --> 00:20:09.990 align:start position:0% there is a risk of overfitting. SVM models demonstrate <00:20:07.811><c>robust </c><00:20:08.102><c>performance </c><00:20:08.393><c>in </c><00:20:08.684><c>non-linear </c><00:20:08.975><c>regression </c><00:20:09.266><c>problems </c><00:20:09.557><c>and</c> 00:20:09.990 --> 00:20:10.000 align:start position:0% demonstrate robust performance in non-linear regression problems and 00:20:10.000 --> 00:20:12.950 align:start position:0% demonstrate robust performance in non-linear regression problems and can <00:20:10.394><c>achieve </c><00:20:10.788><c>good </c><00:20:11.182><c>results </c><00:20:11.576><c>on </c><00:20:11.970><c>relatively </c><00:20:12.364><c>small </c><00:20:12.758><c>datasets.</c> 00:20:12.950 --> 00:20:12.960 align:start position:0% can achieve good results on relatively small datasets. 00:20:12.960 --> 00:20:16.549 align:start position:0% can achieve good results on relatively small datasets. However, <00:20:14.000><c>hyperparameter </c><00:20:15.040><c>tuning </c><00:20:16.080><c>is</c> 00:20:16.549 --> 00:20:16.559 align:start position:0% However, hyperparameter tuning is 00:20:16.559 --> 00:20:19.510 align:start position:0% However, hyperparameter tuning is critical, <00:20:16.775><c>and </c><00:20:16.991><c>training </c><00:20:17.207><c>times </c><00:20:17.423><c>may </c><00:20:17.639><c>be </c><00:20:17.855><c>long </c><00:20:18.071><c>when </c><00:20:18.287><c>the </c><00:20:18.503><c>data </c><00:20:18.719><c>size </c><00:20:18.935><c>is </c><00:20:19.151><c>large.</c> 00:20:19.510 --> 00:20:22.270 align:start position:0% critical, and training times may be long when the data size is large. 00:20:22.270 --> 00:20:22.280 align:start position:0% 00:20:22.280 --> 00:20:24.430 align:start position:0% In <00:20:22.347><c>this </c><00:20:22.414><c>project, </c><00:20:22.481><c>we </c><00:20:22.548><c>compared </c><00:20:22.615><c>and </c><00:20:22.682><c>analyzed </c><00:20:22.749><c>various </c><00:20:22.816><c>models </c><00:20:22.883><c>with </c><00:20:22.950><c>the </c><00:20:23.017><c>goal </c><00:20:23.084><c>of </c><00:20:23.151><c>operational </c><00:20:23.218><c>optimization </c><00:20:23.285><c>by </c><00:20:23.352><c>precisely </c><00:20:23.419><c>forecasting </c><00:20:23.486><c>the </c><00:20:23.553><c>demand </c><00:20:23.620><c>and </c><00:20:23.687><c>inventory </c><00:20:23.754><c>of </c><00:20:23.821><c>corporate </c><00:20:23.888><c>products. </c><00:20:23.955><c>As </c><00:20:24.022><c>a</c> 00:20:24.430 --> 00:20:27.430 align:start position:0% In this project, we compared and analyzed various models with the goal of operational optimization by precisely forecasting the demand and inventory of corporate products. As a 00:20:27.430 --> 00:20:30.149 align:start position:0% 00:20:30.149 --> 00:20:30.159 align:start position:0% 00:20:30.159 --> 00:20:33.149 align:start position:0% result, <00:20:32.000><c>the</c> 00:20:33.149 --> 00:20:33.159 align:start position:0% result, the 00:20:33.159 --> 00:20:36.950 align:start position:0% result, the ensemble <00:20:33.685><c>model </c><00:20:34.211><c>combining </c><00:20:34.737><c>TFT, </c><00:20:35.263><c>LSTM, </c><00:20:35.789><c>and </c><00:20:36.315><c>XGB</c> 00:20:36.950 --> 00:20:36.960 align:start position:0% ensemble model combining TFT, LSTM, and XGB 00:20:36.960 --> 00:20:39.590 align:start position:0% ensemble model combining TFT, LSTM, and XGB demonstrated <00:20:37.340><c>the </c><00:20:37.720><c>best </c><00:20:38.100><c>performance </c><00:20:38.480><c>and </c><00:20:38.860><c>recorded </c><00:20:39.240><c>high</c> 00:20:39.590 --> 00:20:39.600 align:start position:0% demonstrated the best performance and recorded high 00:20:39.600 --> 00:20:41.110 align:start position:0% demonstrated the best performance and recorded high prediction <00:20:39.960><c>accuracy.</c> 00:20:41.110 --> 00:20:41.120 align:start position:0% prediction accuracy. 00:20:41.120 --> 00:20:43.590 align:start position:0% prediction accuracy. Through <00:20:41.394><c>this, </c><00:20:41.668><c>we </c><00:20:41.942><c>derived </c><00:20:42.216><c>meaningful </c><00:20:42.490><c>results </c><00:20:42.764><c>confirming </c><00:20:43.038><c>the</c> 00:20:43.590 --> 00:20:43.600 align:start position:0% Through this, we derived meaningful results confirming the 00:20:43.600 --> 00:20:46.190 align:start position:0% Through this, we derived meaningful results confirming the potential <00:20:43.826><c>for </c><00:20:44.052><c>practical </c><00:20:44.278><c>application </c><00:20:44.504><c>in </c><00:20:44.730><c>demand </c><00:20:44.956><c>forecasting </c><00:20:45.182><c>and </c><00:20:45.408><c>inventory </c><00:20:45.634><c>management.</c> 00:20:46.190 --> 00:20:48.870 align:start position:0% potential for practical application in demand forecasting and inventory management. 00:20:48.870 --> 00:20:48.880 align:start position:0% 00:20:48.880 --> 00:20:52.880 align:start position:0% Thank <00:20:49.880><c>you.</c>