VIETNAM NATIONAL UNIVERSITY

UNIVERSITY OF INFORMATION TECHNOLOGY

INFORMATION SYSTEMS FACULTY

DEMAND FORECASTING WITH

MACHINE LEARNING
COURSE: Supply Chain Management
Lecturer : Dr. Tran Van Hai Trieu
Class : [Link]
: Pham Thi Thuy Trang - 21522697
: Dang Huu Quang - 21522508
Group 1
: Huynh Thi Ha Giang - 21522021
: Vu Bao Han - 21522040

Ho Chi Minh City, October 2024

Supply Chain - Demand Forecasting with Machine Learning Group 1

ACKNOWLEDGEMENTS
For the beginning, we would like to send our warmest thank you to all the faculty
from University of Information and Technology for providing us all the fundamental
knowledge to implement this project.
Nevertheless, for us to develop this project, it is impossible not to mention
[Link] Van Hai Trieu. Thank you for your devotional lessons and valuable sharing to
help us gain more knowledge about Supply Chain Management.
We had done this project in more than a month, and from the bottom of our heart,
that was a precious moment in our student life. This project helped us to engage
ouranalysis, development, processing abilities and especially collaboration skills. We
have had a clearer aspect about our future career path. This is also contributing to our
life experience package that will always go alongside us until the end of our life.
Based on all the fundamental studies combined with enthusiastic support from
the instructors, we have completed this project successfully. However, since this is our
very first project, it is impossible to avoid some minor mistakes.
We are looking forward to receiving valuable comments and advice from all
lecturers to improve our wider scale project soon.
Once again, we would like to sincerely send to you a huge thanks from the
bottom of our heart.

Ho Chi Minh city, November 2024

2
Supply Chain - Demand Forecasting with Machine Learning Group 1

LECTURER’S FEEDBACK
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................

Ho Chi Minh City, October 28, 2024

3
Supply Chain - Demand Forecasting with Machine Learning Group 1

TABLE OF CONTENTS
ACKNOWLEDGEMENTS ................................................................................. 2
LECTURER’S FEEDBACK ............................................................................... 3
TABLE OF CONTENTS .................................................................................... 4
LIST OF FIGURES ............................................................................................ 8
LIST OF TABLES ............................................................................................... 9
I. GENERAL INTRODUCTION: ................................................................. 10
1. Supply chain project overview: ............................................................ 10
2. Project scope:.......................................................................................... 10
3. Project objectives: .................................................................................. 10
4. Master data:............................................................................................ 11
II. RELATED WORDS: ............................................................................... 22
III. BUSINESS PROCESS SYSTEM: .......................................................... 23
1. Overview of Walmart: ........................................................................... 23
2. Introduction to the supply chain strategy of Walmart:...................... 23
3. Business strategy: ................................................................................... 24
4. Supply Chain Management ................................................................... 26
4.1. Global sourcing, in-store warehousing and cross-docking: ........... 26
4.2. Technology and Automation Integrated: ......................................... 27
4.3. Workforce Management as third largest employer: ....................... 28
4.4. Conclusion: ......................................................................................... 28
5. Competitive Advantage: ........................................................................ 29
6. Value Created:........................................................................................ 31
7. Customer segments: ............................................................................... 31
8. Sales Channel: ........................................................................................ 34
8.1. Physical Stores: ................................................................................... 34
8.2. E-commerce: .................................................................................... 34
8.3. Marketplace: .................................................................................... 35
8.4. Mobile Sales: .................................................................................... 35
8.5. Membership Program: ................................................................... 35
8.6. Direct Sales: ..................................................................................... 35
8.7. Omnichanel:..................................................................................... 35
IV. MODELING:........................................................................................... 36

4
Supply Chain - Demand Forecasting with Machine Learning Group 1

1. Decision Tree: ......................................................................................... 36

2. Support Vector Machine (SVM): ......................................................... 37
3. Random Forest: ...................................................................................... 38
V. EVALUATION METRICS: ....................................................................... 39
1. R-squared (R²):....................................................................................... 39
2. Mean Absolute Error (MAE): .............................................................. 39
3. Mean Squared Error (MSE): ................................................................ 40
4. Root Mean Squared Error (RMSE): .................................................... 40
VI. RESULT: ................................................................................................. 40
VII. CONCLUSION AND SUGGESTION FOR FUTURE DIRECTIONS:
42
REFERENCES.................................................................................................. 45

5
 Supply Chain - Demand Forecasting with Machine Learning Group 1

ASSIGNMENT AND MEMBER EVALUATION

TABLE
Name Student ID Assignment Evaluation
• Plan for project, divide tasks and
timeline for project
• Check the data after teammate
research.
• Research the contents of project
Pham Thi Thuy 21522697
to report. 100%
Trang (Leader) • Make the conclusion and
suggestion for supply chain.
• Do the report.
• Do the PowerPoint.
• Presenter.

• Research supply chain process

of Walmart.

Vu Bao Han 21522040 • Do the report. 100%

• Do the PowerPoint.
• Presenter.

• Preprocessing data.
• Run models to predict future.
Huynh Thi Ha
21522021 • Do the report. 100%
Giang
• Do the PowerPoint.
• Presenter.

• Research and describe EDA of

the data.
Dang Huu Quang 21522508 100%
• Do the report.
• Do the PowerPoint.

6
Supply Chain - Demand Forecasting with Machine Learning Group 1

• Presenter.

7
Supply Chain - Demand Forecasting with Machine Learning Group 1

LIST OF FIGURES

Figure 1: EDA with non scale data .................................................................... 13

Figure 2: Distribution, spread and presence of outliers for each feature ........... 15
Figure 3: Correlation matrix for various features ............................................... 17
Figure 4: EDA with scale data ........................................................................... 18
Figure 5: Distribution, spread and presence of outliers for each feature ........... 19
Figure 6: Correlation for various features .......................................................... 21
Figure [Link] worldwide from fiscal year 2012 to 2024 ............................. 26
Figure [Link] retailers index with shoppers with age group ............................ 34
Figure 9: Decision tree model ............................................................................ 36
Figure 10: Support Vector Machine (SVM) model............................................ 37
Figure 11: Random Forest model ....................................................................... 39

8
Supply Chain - Demand Forecasting with Machine Learning Group 1

LIST OF TABLES
Table 1: Data description ................................................................................... 11
Table 2: Customer segments .............................................................................. 32
Table 3: Result for exogenous(external) variables ............................................. 40
Table 4: Result for internal variables ................................................................. 41
Table 5: Result for all variables ......................................................................... 41

9
Supply Chain - Demand Forecasting with Machine Learning Group 1

I. GENERAL INTRODUCTION:
In this first chapter, we will discuss about the project overview, reason to
choose this topic, scopes, objectives, and master data.

1. Supply chain project overview:

Forecasting demand for Walmart products is a critical task that can significantly
impact the retailer's supply chain efficiency, inventory management, and overall
profitability. By accurately predicting future demand, Walmart can optimize its
inventory levels, reduce stockouts and overstock, and improve customer satisfaction.

Reason to choose this topic:

• Real-world Impact: Accurate demand forecasting can directly impact

Walmart's bottom line by reducing costs and increasing revenue.
• Data Availability: Walmart provides a publicly accessible dataset
containing historical sales data, which is essential for training and
evaluating machine learning models.
• Machine Learning Advancement: Recent advancements in machine
learning techniques have made it possible to develop sophisticated models
for time series forecasting and demand prediction.

2. Project scope:

This project focuses on applying machine learning techniques to accurately

predict product demand through historical sales data from Walmart. The main goal is
to harness data and predictive models to capture customer demand in stores, thereby
optimizing supply chain processes, helping to increase profits, improve customer
satisfaction, and drive sustainable sales in the future. Through a detailed analysis of the
process, assessment of existing strengths and weaknesses, the project aims to propose
solutions to overcome and improve the limitations, help the supply chain operate more
efficiently and better meet the needs of the market.

3. Project objectives:

Using machine learning to predict customers demand at Walmart Store through

Weekly Sale data to:
• Accurate Demand Forecasting: Develop a machine learning model that can
accurately predict future demand for Walmart products.
• Optimized Inventory Management: Use accurate demand forecasts to
optimize inventory levels, reducing stockouts and overstock.
• Improved Decision Making: Provide valuable insights to Walmart's
decision-makers to support strategic planning and operational efficiency.

10
Supply Chain - Demand Forecasting with Machine Learning Group 1

• Enhanced Customer Satisfaction: Ensure product availability by

accurately predicting demand and optimizing supply chain operations.

4. Master data:1
Dataset: Historical sales data for 45 Walmart stores located in different regions are
available in US from 5-2-2010 to 26-10-2012.

Case study:

One of the leading retail stores in the US, Walmart, would like to predict the
sales and demand accurately. There are certain events and holidays which impact
sales on each day. There are sales data available for 45 stores of Walmart. The
business is facing a challenge due to unforeseen demands and runs out of stock some
times, due to the inappropriate machine learning algorithm. An ideal ML algorithm
will predict demand accurately and ingest factors like economic conditions
including CPI, Unemployment Index, etc.

In order to increase supply resources, Walmart runs several promotional

markdown events throughout the year. These markdowns precede prominent
holidays, the four largest of all, which are the Super Bowl, Labour Day,
Thanksgiving, and Christmas. The weeks including these holidays are weighted five
times higher in the evaluation than non-holiday weeks. Then the question is how
does it impact the whole supply chain of this company? We will study the sales data
of one of the largest retailers in the world. Let's figure out what factors influence its
revenue. [8] Can factors such as air temperature and fuel cost influence the success
of a huge company along with the purchasing power index and seasonal discounts?
And how does machine learning minimize costs and increase economic impact?
Let’s look deep into.

Datasets describe:

Data consists of 421570 records of weekly sales from stores spanning between
‘05-Feb-2010’ to ’26-Oct-2012’. This comprises of 143 Weeks of sales data. Within
this file you will find the following fields:

Table 1: Data description

Field Name Description

Store The store number on the whole US. (using number to make
represent for area store located)
Date Sales for the given store (Date)

1
H, M. Y. (2021, December 26). Walmart dataset. Kaggle. [Link]
dataset

11
Supply Chain - Demand Forecasting with Machine Learning Group 1

Weekly_Sales The revenue for sales open at a store. (dollar)

Holiday_ Flag Determinate whether the week is a special holiday (1 –
Holiday and 0 – Non-holiday week)
Temperature Temperature on the day of sale. (F Temperature)
Fuel_Price Cost of gas fuel in the region which the store located.
(dollar/gallon)
CPI Prevailing Consumer Price Index (%).
Unemployment Prevailing unemployment rate in US at the period time (%)

The Holiday Events includes:

• Super Bowl: 12-Feb-10, 11-Feb-11, 10-Feb-12, 8-Feb-13

• Labour Day: 10-Sep-10, 9-Sep-11, 7-Sep-12, 6-Sep-13

• Thanksgiving: 26-Nov-10, 25-Nov-11, 23-Nov-12, 29-Nov-13

• Christmas: 31-Dec-10, 30-Dec-11, 28-Dec-12, 27-Dec-13.

Supply Chain Problems Approach:

Several approach way to get data insight are finding the reasonable answer or the
convince hypothesis for these questions below:

Is there any difference between normal day with holiday weeks their overall sales
volume across different regions?

 We may get the insights here that can help decide if more resources need to
be allocated to specific regions or stores during holiday periods.

What is the fuel price changes, CPI, temperature appear in here? Is there a significant
correlation between these factors and sales volumes?

 Understanding this relationship could help in planning supply routes and

managing logistics costs when fuel prices are high.

Holiday? Everybody knows the super-massive demand in this period of time but
what is the ROI (Return on investment) of holiday-related promotions, and how can
it be optimized?

 By analyzing sales growth around holidays, you can calculate the ROI and
optimize promotional efforts for future holiday seasons.

45 Stores distributed on whole US so there are required a large chain of supply and
provide on time this make us consider What are the inventory turnover rates by store

12
Supply Chain - Demand Forecasting with Machine Learning Group 1

and season? and how does temperature affect sales across different geographic
regions?

 Warmer or colder temperatures could influence sales patterns (e.g., cold

weather items may spike in winter). Understanding this allows for temperature-
sensitive stocking decisions.

All considerations above are good way to approach, let's see what about our team
solutions.

Data insight:

1. EDA with non-scale data:

Figure 1: EDA with non scale data

Weekly Sales: [first row- on left]

• This histogram is right-skewed, indicating that most stores have weekly sales
in the lower range, with a few instances of high sales.
• A large proportion of weekly sales values fall under 1 million, while sales
above 2 million are rare.

Holiday Flag: [first row- on right]

13
Supply Chain - Demand Forecasting with Machine Learning Group 1

• Simply remark the number of weeks have holiday.

Temperature: [second row- on left]

• The temperature distribution is roughly bell-shaped, suggesting a normal

distribution centered around 60–70 degrees Fahrenheit.
• Most temperatures are between 40 and 80 degrees, with fewer records at the
extremes.

Fuel Price: [second row- on right]

• This distribution has multiple peaks, indicating variability in fuel prices over
time.
• A large number of fuel prices cluster around 3.5 dollar/gallon, but there are
notable groups at other prices, suggesting fluctuation.

CPI (Consumer Price Index): [last row- on left]

• The CPI distribution shows a bimodal pattern with two main clusters: one
around 130–140 and another around 200–220.
• This indicates that CPI has varied over the period covered by the dataset,
potentially due to economic shifts.

Unemployment: [last row- on right]

• The unemployment distribution is skewed to the right, with most values

concentrated around 6–8%.

These insights suggest different ranges and distributions for each feature, which could
impact how they relate to weekly sales in further analysis, such as through correlation
or regression.

14
Supply Chain - Demand Forecasting with Machine Learning Group 1

Figure 2: Distribution, spread and presence of outliers for each feature

The boxplots provide insights into the distribution, spread, and presence of
outliers for each feature in the dataset. We are going to detailed insights for each plot:

1. Weekly Sales [first row-the left]

• The boxplot shows that the majority of weekly sales are below 1.5 million,
with the median slightly above 1 million.
• There are several outliers on the higher end, with weekly sales exceeding 2
million, which indicates that while most sales are within a common range, a
few stores have notably high weekly sales.

2. Holiday Flag [first row-the middle]

• The boxplot for the Holiday Flag is binary, with values of either 0 or 1.

3. Temperature: [first row-the right]

• The temperature distribution is fairly centered, with most values between 40-
and 80-degrees Fahrenheit.
• There are some outliers on both lower and upper ends, representing instances
of extreme temperatures.
• The median temperature is around 60 degrees, with a relatively symmetrical
spread around this point.

4. Fuel Price: [second row-the left]

15
Supply Chain - Demand Forecasting with Machine Learning Group 1

• The fuel price data has a moderately widespread, with most values ranging
from approximately 2.75 to 4.00.
• There are no significant outliers, suggesting that fuel prices remained within a
consistent range over the recorded period.
• The median fuel price is close to 3.25, and the distribution appears relatively
balanced.

5. CPI (Consumer Price Index):

• The CPI values are distributed with a slight skew, with most values around the
median, approximately 180.
• There are a few outliers below 140 and above 220, which indicate extreme CPI
values during the period covered by the dataset, possibly due to economic
events.
• The middle 50% of values are spread over a broad range, showing variability
in CPI over time.

6. Unemployment:

• Unemployment data has a wider range of outliers, particularly on the higher

end, where unemployment exceeds 10%.
• The majority of values lie between 6% and 8%, with a median close to 7%,
suggesting that unemployment remained relatively stable for most of the
dataset.

In summary, these boxplots reveal the central tendencies and variability of each
feature, along with the presence of outliers that may indicate unique events or
conditions, such as high sales weeks, holiday effects, extreme temperatures, or
economic shifts. These outliers could be points of interest for further analysis or
modeling, as they might have a unique impact on weekly sales patterns.

The heatmap shows the correlation matrix for various features in the dataset,
indicating how each variable is linearly related to the others:

16
Supply Chain - Demand Forecasting with Machine Learning Group 1

Figure 3: Correlation matrix for various features

• Weekly Sales: Has a weak negative correlation with Store (-0.34), meaning
stores with higher IDs tend to have slightly lower weekly sales; Very weak
correlations with other variables, suggesting sales aren’t strongly affected by
factors like holiday weeks, fuel price, temperature, CPI, or unemployment.

• Holiday Flag: Has weak negative correlations with Temperature (-0.16) and
Fuel Price (-0.078), indicating that holiday weeks may slightly coincide with
lower temperatures and fuel prices.

• Temperature: Shows weak positive correlations with Fuel Price (0.14) and
CPI (0.18), possibly reflecting seasonal effects.

• Unemployment: Negatively correlated with CPI (-0.3), suggesting that higher

unemployment is associated with lower CPI values. Slight positive correlation
with Store (0.22), indicating regional or temporal variations in unemployment
by store location.

Overall, the correlations are weak, indicating that these variables are mostly
independent or have minimal linear relationships.

17
Supply Chain - Demand Forecasting with Machine Learning Group 1

2. EDA with scale data:

Figure 4: EDA with scale data

Variable: Weekly Sales [First Row – Left]

• Distribution: The distribution is right-skewed, meaning there are a few very

high sales values and most of the values are concentrated towards the lower
end. This indicates that there might be some outliers or extreme sales figures.

Variable: Holiday Flag [First Row – Right]

• The plot for the Holiday Flag is binary, with values of either 0 or 1.

Variable: Temperature [Second Row – Left]

• Distribution: The distribution is approximately normal, with a slight right

skew. This suggests that the temperature values are centered around a certain point,
and there are some higher temperatures.

Variable: Fuel Price [Second Row – Right]

18
Supply Chain - Demand Forecasting with Machine Learning Group 1

• Distribution: The distribution is approximately normal, with a slight right

skew. This suggests that the fuel prices are centered around a certain point, and
there are some higher fuel prices.

Variable: CPI: [Last Row – Left]

• Distribution: The distribution is right-skewed, meaning there are a few very

high CPI values and most of the values are concentrated towards the lower
end. This indicates that there might be some outliers or extreme CPI figures.

Variable: Unemployment [Last Row – Right]

• Distribution: The distribution is approximately normal, with a slight right

skew. This suggests that the unemployment rates are centered around a certain
point, and there are some higher unemployment rates.

Figure 5: Distribution, spread and presence of outliers for each feature

Weekly Sales [First Row – Left]:

• The box plot shows that the median weekly sales are around 0.4. The box
indicates the interquartile range (IQR), which is the range of the middle 50%
of the data. The whiskers extend to the minimum and maximum values within
a certain range (usually 1.5 times the IQR). The dots beyond the whiskers
represent outliers. This suggests that there are a few stores with significantly
higher weekly sales compared to the majority.

Holiday Flag [First Row – Middle]

19
Supply Chain - Demand Forecasting with Machine Learning Group 1

• The boxplot for the Holiday Flag is binary, with values of either 0 or 1.

Temperature [First Row – Right]

• The box plot shows that the median temperature is around 0.6. The box
indicates the interquartile range (IQR). The whiskers extend to the minimum
and maximum values within a certain range. The dots beyond the whiskers
represent outliers. This suggests that there are a few stores with significantly
higher or lower temperatures compared to the majority.

Fuel Price [Second Row – Left]

• The box plot shows that the median fuel price is around 0.4. The box indicates
the interquartile range (IQR). The whiskers extend to the minimum and
maximum values within a certain range. There are no outliers visible in this
box plot.

CPI [Second Row – Middle]

• The box plot shows that the median CPI is around 0.4. The box indicates the
interquartile range (IQR). The whiskers extend to the minimum and maximum
values within a certain range. There are no outliers visible in this box plot.

Unemployment [Second Row – Right]

• The box plot shows that the median unemployment rate is around 0.4. The box
indicates the interquartile range (IQR). The whiskers extend to the minimum
and maximum values within a certain range. The dots beyond the whiskers
represent outliers. This suggests that there are a few stores with significantly
higher unemployment rates compared to the majority.

20
Supply Chain - Demand Forecasting with Machine Learning Group 1

Figure 6: Correlation for various features

Key Insights from the Heatmap:

Strong Positive Correlations:

• Store and Fuel Price: This indicates that stores with higher fuel prices tend to
have higher overall sales.
• Store and CPI: This suggests that stores in areas with higher Consumer Price
Index (CPI) generally have higher sales.

Moderate Positive Correlations:

• Store and Temperature: There's a moderate positive relationship between store

sales and temperature. This could be due to various factors like seasonal
shopping trends or increased foot traffic in warmer weather.

Weak Negative Correlations:

21
Supply Chain - Demand Forecasting with Machine Learning Group 1

• Store and Holiday Flag: This suggests that stores might see slightly lower sales
during holidays, possibly due to reduced business hours or customer
preferences.
• Store and Weekly Sales: This indicates a very weak negative correlation,
which might not be practically significant.

No Significant Correlations:

• Store and Unemployment: This suggests that unemployment rates in the area
don't have a strong impact on store sales.

II. RELATED WORDS:

1. Demand Forecasting in Supply Chain Using Uni-Regression Deep Approximate

Forecasting Model (Emad Aldahmani, Ahmad Alzubi * and Kolawole Iyiola)2

This research presents a uni-regression deep approximate forecasting model for

predicting future demand in supply chains, tackling issues like complex patterns,
external factors, and nonlinear relationships. It diverges from traditional models by
employing a deep learning strategy through recurrent bidirectional long short-term
memory (BiLSTM) and nonlinear autoregressive with exogenous inputs (NARX),
focusing on regression-based approaches. The model can capture intricate
dependencies and patterns that elude conventional approaches. The integration of
BiLSTM and NARX provides a robust foundation for accurate demand forecasting.
The novel uni-regression technique significantly improves the model’s capability to
detect intricate patterns and dependencies in supply chain data, offering a new angle
for demand forecasting. This approach not only broadens the scope of modeling
techniques but also underlines the value of deep learning for enhanced accuracy in
the fluctuating supply chain sector. The uni-regression model notably outperforms
existing models in accuracy, achieving the lowest errors: mean average error (MAE)
at 1.73, mean square error (MSE) at 4.14, root mean square error (RMSE) at 2.03,
root mean squared scaled error (RMSSE) at 0.020, and R-squared at 0.94. This
underscores its effectiveness in forecasting demand within dynamic supply chains.
Practitioners and decision-makers can leverage the uni-regression model to make
informed decisions, optimize inventory management, and enhance supply chain
resilience. Furthermore, the findings contribute to the ongoing evolution of supply
chain demand forecasting methodologies.

2. Machine learning demand forecasting and supply chain performance (Javad

Feizabadi):3

2
Aldahmani, E., Alzubi, A., & Iyiola, K. (2024). Demand Forecasting in Supply Chain Using Uni-Regression
Deep Approximate Forecasting Model. Applied Sciences, 14(18), 8110. [Link]
3
Feizabadi, J. (2020). Machine learning demand forecasting and supply chain performance. International
Journal of Logistics Research and Applications, 25(2), 1–24.

22
Supply Chain - Demand Forecasting with Machine Learning Group 1

This research paper presents a hybrid approach to demand forecasting that

combines time series models with leading indicators and machine learning
techniques. The study found that this hybrid approach significantly improves
forecasting accuracy compared to traditional methods. By incorporating machine
learning, the model can capture complex relationships between various factors
influencing demand, leading to more accurate predictions. The research highlights
the potential benefits of using machine learning for demand forecasting in supply
chain management and provides valuable insights for practitioners.

3. Forecasting Supply Chain Demand Using Machine Learning Algorithms (Réal

André Carbonneau, Rustam Vahidov, Kevin Laframboise):4

This research paper investigates the effectiveness of machine learning

techniques in forecasting demand signals with high noise-to-pattern ratios within
supply chain contexts. While traditional methods like exponential smoothing often
outperform advanced machine learning techniques for single-product forecasting,
the study found that combining data from multiple products and using a Support
Vector Machine (SVM) can significantly improve accuracy. This "Super Wide"
approach, while computationally intensive, can lead to substantial cost savings by
reducing inventory errors.

III. BUSINESS PROCESS SYSTEM:

1. Overview of Walmart:5

Walmart is a retailer that operates grocery stores, supermarkets, hypermarkets,

department and discount stores, and neighborhood markets. It has grown to be the
largest and probably most powerful retailer in the world, with sales, inventory turnover,
and operating profit exceeding those of any other retailer in recent decades. To
understand the performance of Walmart and its impact on the retail sector, look at its
historical stock figures. Corporate success involves successful supply chain
management as it transitions from a localized store to a worldwide retail organization.
According to Supply Chain Digest, Walmart runs over 11,700 shops under 59 brand
names, employs 2.3 million people in 28 countries, and manages a $32 billion inventory
on average.

2. Introduction to the supply chain strategy of Walmart:6

Walmart's supply chain management strategy has given the corporation various
long-term competitive benefits, such as lower product prices, lower inventory carrying

4
Carbonneau, R., Vahidov, R., & Laframboise, K. (2007). Machine Learning-Based Demand Forecasting in
Supply Chains. International Journal of Intelligent Information Technologies, 3(4), 40–57.
[Link]
5
GlobalData. (2022). Company Profile. GlobalData. [Link]
inc/
6
Rickerby, M. (2020, January 4). Walmart supply chain: Winning at inventory logistics. Extensiv.
[Link]

23
Supply Chain - Demand Forecasting with Machine Learning Group 1

costs, more in-store diversity and selection, and extremely competitive customer
pricing.

Walmart's ability to sell items at a lower price than competitors is dependent on

reducing the number of intermediaries. They always try to buy directly from the
manufacturer. With its massive purchasing power, Walmart uses this data collection to
affect the behavior of its suppliers, lowering prices dramatically.

Walmart also uses discount programs on a regular basis to encourage effective

consumption, ensuring that there is no inventory at the end of the year.
This is corroborated by Walmart's sale seasons with almost guaranteed success. Three
reasons helped Wal-Mart establish its supply chain successfully:

1. Distribution Operation Process

2. How to operate fleet and technology
3. Inventory management and product forecasting.

3. Business strategy7:

When it comes to business strategy, Walmart is defined as one of the most

successful business cases, which inspired and provided many business ideas for average
to big scaled retailed start-up and businesses.

• Pricing Strategy

o Walmart is well-known for their EDLP as Everyday Low Prices, this is not
only the slogan to attract the consumers, but the commitment to affordability
is the cornerstone of Walmart’s marketing methods to shape consumers’
perceptions to directly driving foot to their stores.
o At this approach, Walmart did make a big step to a transition to an Omni
Chanel Marketplace. They are not only offer low prices almost every day
for consumers but continuously monitor competitors’ prices then provide a
dynamic pricing to remain their competitivity.

• Supply Chain Management

o Walmart's approach to supply chain management is defined by its focus on

efficiency, the use of advanced technology, robust partnerships with
suppliers, and a dedication to sustainability. By consistently refining its
supply chain processes, Walmart effectively lowers costs while also

7
Choo, J. (2024, January 5). Walmart business strategy: A comprehensive analysis. THE STRATEGY
JOURNEY. [Link]
business-strategy-

24
Supply Chain - Demand Forecasting with Machine Learning Group 1

improving customer satisfaction, which helps reinforce its status as a leader

in the retail sector.

• Product Strategy

o Walmart provides a wide range of products which ensure that customer only
need to visit their place to purchase everything. The diversity in their
product range helps them to ensure the commitment rate of customers and
enhance the sustainability.

• Marketing & Customer Engagement

o Walmart focuses on marketing and customer engagement through several

key strategies. They utilize data-driven marketing to gain insights into
customer preferences and customize promotions accordingly. Additionally,
programs like Walmart+ offer extra benefits to customers, fostering loyalty
and encouraging repeat purchases. Walmart also adopts an omnichannel
approach, its physical stores and online platforms to create a convenient
shopping experience for customers.

• Global Expansion

o Walmart's strategy for global expansion focuses on adapting to local

markets and forming partnerships to effectively navigate the challenges in
emerging economies such as India and Brazil. By prioritizing cultural
understanding, customizing its product offerings, and ensuring efficient
logistics, Walmart seeks to build a solid presence and foster growth in these
important regions.

• Corporate Social Responsibility

o Walmart’s Corporate Social Responsibility (CSR) efforts center around

sustainability, community involvement, and ethical sourcing. The company
is committed to reducing its carbon footprint by improving energy
efficiency and investing in renewable energy sources. It also plays an active
role in supporting local communities through donations, disaster relief
initiatives, and educational programs. Furthermore, Walmart prioritizes
ethical sourcing to ensure that its suppliers adhere to fair labor practices.
Overall, these initiatives are designed to make a positive impact on society
while fostering trust and loyalty among customers and stakeholders.

• Digital Transformation

o Walmart's digital transformation focuses on enhancing customer

experiences and streamlining operations through technology. The company
has invested in e-commerce, offering features like same-day delivery and

25
Supply Chain - Demand Forecasting with Machine Learning Group 1

curbside pickup. It uses data analytics and AI for efficient inventory

management and personalized marketing, while also adopting automation
to reduce costs and improve service speed. Overall, these efforts aim to keep
Walmart competitive in the evolving retail landscape.

4. Supply Chain Management

Figure [Link] worldwide from fiscal year 2012 to 2024

4.1. Global sourcing, in-store warehousing and cross-docking: 8

Walmart utilizes a global sourcing strategy that allows it to source products from
a wide range of suppliers around the world. This extensive supply chain enables the
company to offer competitive prices across a diverse product portfolio that includes
over 18 categories, ranging from groceries to electronics.

To effectively manage this broad assortment of products, many Walmart stores

feature upper-level storage areas for additional inventory. This in-store warehousing
reduces the need for frequent deliveries, helping to keep costs down while ensuring that
shelves are consistently stocked. In addition to in-store storage, Walmart operates a
robust network of distribution centers that are essential for regional logistics, storage,
and packaging of online orders.

8
Lövenich, A. (2024, September 23). The amazing supply chain of Walmart. All Things Supply Chain - Supply
Chain Trends, Best Practices, News and Much More!; All Things Supply Chain.
[Link]

26
Supply Chain - Demand Forecasting with Machine Learning Group 1

Walmart has also adopted new delivery methods to improve efficiency and cost-
effectiveness. A key technique is cross-docking, which eliminates the need for
traditional warehousing by allowing goods to move directly from inbound to outbound
shipments. Suppliers deliver pre-packaged products to distribution centers, where they
are quickly loaded onto trucks that are ready to deliver to stores. This process minimizes
unloading and loading times, ensuring that products reach store shelves more quickly.
By utilizing centralized distribution centers that automatically allocate goods to stores,
Walmart streamlines its operations and reduces overall costs.

 Walmart's supply chain management strategy highlights its focus on efficiency and
cost control. The global sourcing approach not only allows Walmart to keep prices
low but also enhances its ability to offer a vast array of products to consumers. The
incorporation of in-store warehousing optimizes inventory management and
improves customer satisfaction by ensuring product availability.

 The cross-docking technique exemplifies Walmart's innovative logistics practices,

significantly speeding up the supply chain process. By reducing reliance on
extensive warehousing, Walmart minimizes costs and enhances operational
efficiency. This strategic combination of global sourcing, in-store warehousing, and
advanced logistics positions Walmart as a leader in retail, enabling it to respond
swiftly to consumer demands while maintaining a competitive edge in the market.
Overall, Walmart's supply chain strategies reflect a well-coordinated effort to
balance cost efficiency with customer service excellence.

4.2. Technology and Automation Integrated:

Walmart’s strategic embrace of technology and automation has fundamentally

transformed its supply chain operations. By adopting RFID and AI, Walmart has
significantly enhanced its inventory management capabilities, leading to improved
product availability and reduced stockouts. This technological integration not only
streamlines operations but also empowers Walmart to respond swiftly to shifting
consumer demands, which is critical in the highly competitive retail environment.

The substantial investment in automation, particularly the $14 billion spent in

2021, underscores Walmart's commitment to optimizing its supply chain. The
implementation of advanced robotics and automated systems not only increases
efficiency but also minimizes human error, which can be a significant factor in
fulfillment processes. The shift toward automated fulfillment centers illustrates
Walmart's forward-thinking approach, positioning the company to handle growing e-
commerce demands effectively.

Moreover, the use of vendor-managed inventory systems exemplifies Walmart's

focus on collaboration with suppliers. By providing real-time data, Walmart enhances

27
Supply Chain - Demand Forecasting with Machine Learning Group 1

supplier relationships, reduces excess inventory, and ensures that shelves remain
stocked. This collaborative strategy not only drives down costs but also enhances
customer satisfaction by ensuring product availability.

Overall, Walmart's integration of technology and automation not only improves

operational efficiency but also strengthens its competitive advantage in the retail
market. The company's ability to adapt to technological advancements while
maintaining strong supplier relationships and focusing on customer needs positions it
well for future growth and success in an increasingly digital landscape.

4.3. Workforce Management as third largest employer:

Walmart's approach to workforce management is a strategic effort to optimize

its vast employee base while enhancing job satisfaction. By offering diverse career
paths and utilizing technology like the Me@Walmart app, this company addresses the
varying needs of its employees, which is critical in retaining talent in a competitive job
market.

The app's features, such as shift visibility and mobile clock-in, illustrate
Walmart's understanding of modern workforce dynamics, where flexibility and ease of
communication are paramount. This technological investment not only streamlines
operations but also empowers employees to take control of their schedules and work
environment.

Furthermore, the emphasis on communication through the app can lead to a more
engaged workforce. When employees can easily connect with one another, it fosters a
sense of community and teamwork, which is essential for maintaining high levels of
service in retail operations.

Overall, Walmart's workforce management strategies reflect a commitment to

innovation and employee well-being. By effectively leveraging technology and
providing diverse career opportunities, Walmart positions itself as a desirable employer,
capable of attracting and retaining a large, skilled workforce essential for its continued
success in the retail industry.
4.4. Conclusion: 9
Walmart's supply chain management has evolved significantly from 2012 to the
present, focusing on efficiency and technology integration. There are six keys’
developments include:

9
Thi Thu Ha, N. (2017). Wal-Mart’s Successfully Integrated Supply Chain and the Necessity of Establishing the
Triple-A Supply Chain in the 21st Century. ResearchGate.
[Link]

28
Supply Chain - Demand Forecasting with Machine Learning Group 1

a. Distribution Practices: Walmart has enhanced its distribution strategies,

including the use of cross-docking, which minimizes storage time and speeds up
the flow of goods from suppliers to stores. This method allows for quicker
replenishment and reduces inventory costs.
b. Technology Integration: The implementation of advanced technologies such as
RFID (Radio Frequency Identification) has been a major focus. Although the
initial RFID case tagging initiative faced challenges, Walmart shifted towards
item-level tagging for better inventory management and tracking.
c. Sustainability Initiatives: In recent years, Walmart has prioritized sustainability
within its supply chain. This includes the development of a supplier
sustainability index to evaluate and improve the environmental impact of its
vendors.
d. Collaborative Efforts: Walmart has engaged in collaborative planning,
forecasting, and replenishment (CPFR) with suppliers to enhance supply chain
efficiency. This approach allows for better alignment of inventory levels with
consumer demand.
e. Local Sourcing: The company has made significant investments in sourcing
local produce, aiming to reduce transportation costs and support local
economies. This strategy not only improves freshness but also enhances
Walmart's image as a community-focused retailer.
f. E-commerce Adaptation: With the rise of online shopping, Walmart has
adapted its supply chain to support e-commerce operations. This includes the
"Site-to-Store" program and direct-to-consumer shipping, which have become
essential for meeting customer expectations in a digital marketplace.
g. Data Analytics: Walmart leverages big data analytics to optimize its supply
chain operations. By analyzing sales trends and inventory levels, the company
can make informed decisions that enhance efficiency and reduce waste.

Through these developments, Walmart has established itself as a leader in supply

chain management, continuously adapting to market changes and consumer needs
while maintaining its commitment to cost efficiency and customer satisfaction.

5. Competitive Advantage:

Walmart's primary competitive advantage stems from its vision of offering low-
cost goods. In addition to this core strategy, several other factors contribute to Walmart's
competitive edge, which are outlined below:
• Selling Low-Priced Goods
Walmart's central competitive advantage lies in its ability to sell products
at prices lower than those of its competitors. The company operates under the
philosophy of “Everyday low prices,” with the slogan, “We save people’s money
so that they can live better.” Since its inception, Walmart has consistently fulfilled

29
Supply Chain - Demand Forecasting with Machine Learning Group 1

this promise by providing essential retail goods at more affordable prices than
other market players.

• Exceptional Customer Service

Walmart is dedicated not only to offering low prices but also to delivering
excellent customer service. Policies such as a 90-day return policy without
receipts and the use of digital tools to enhance the shopping experience exemplify
this commitment. The company invests in hiring and training programs to ensure
employees are motivated to provide top-notch service to customers.

• Cost Leadership
Walmart can sell goods at such low prices while still making a profit. The
key answer lies in its ability to keep costs low through various resources, including
human capital, knowledge, and large-scale operations. Walmart invests
significantly in technology, which helps reduce expenses. Strategic store
locations, effective management, and technology utilization enable Walmart to
achieve economies of scale and maintain cost leadership in the retail sector.

• Superior Supply Chain Management

Walmart's exceptional supply chain management is one of its most
significant competitive advantages. The company has made substantial
investments in developing its supply chain and warehouse systems. By
continuously implementing technological solutions to enhance efficiency,
Walmart has become a leader in the industry, utilizing barcodes and RFID
technology ahead of its competitors. The establishment of regional warehouses
and distribution centers in strategic locations allows for direct shipping from
manufacturers, significantly lowering transportation, labor, and storage costs. This
efficient planning and management not only reduce inventory costs but also
improve customer satisfaction. Additionally, Walmart focuses on building long-
term relationships with its suppliers.

• Global Expansion
Starting with just one store, Walmart has expanded its operations to 28
countries, with approximately 11,700 stores worldwide. The company’s rapid
penetration into various markets is unmatched, giving it a global presence that
competitors find difficult to replicate.

• Market Dominance in Smaller Communities

A strategic move for Walmart has been to establish a presence in smaller
cities and towns, which typically lack the capacity for multiple retail stores.
Through effective promotions, cost efficiency, and service offerings, Walmart
successfully operates in these areas, making it challenging for other retailers to
enter the market. This strategy has allowed Walmart to achieve a dominant
position in these smaller locations, further solidifying its competitive advantage.

30
Supply Chain - Demand Forecasting with Machine Learning Group 1

 In summary, Walmart's ability to offer low prices, exceptional customer service, cost
leadership, superior supply chain management, global expansion, and market
dominance in smaller communities collectively contribute to its strong competitive
position in the retail industry.

6. Value Created:

Walmart's core values are centered around a commitment to putting customers

first and fostering a culture that prioritizes frontline employees. These principles
guide the company's operations and help it stay aligned with its mission and vision.
Walmart’s core values:

• Customer Focus: At the heart of Walmart's philosophy is a dedication to

understanding and meeting customer needs. This focus is evident in every
aspect of the business, ensuring that customer satisfaction is a top priority.
• Prioritize feedback: Walmart values feedback from its customers and
employees, using it to improve services and products continually. This open
line of communication helps the company stay attuned to what its customers
want.
• Inclusivity, Innovation, and Agility: Walmart strives to be inclusive and
adaptable, embracing new ideas and technologies to enhance the shopping
experience. They are committed to being innovative while maintaining high
standards of service quality.
• Exceptional Performance: Walmart aims for excellence in everything it does,
from customer service to operational efficiency. This commitment to high
performance helps the company stand out in a competitive retail landscape.
• Responsibility: They take their role in the community seriously, striving to act
responsibly and ethically in all its dealings. This includes sustainable practices
and a commitment to social responsibility.
• Honesty and Fairness: Integrity is a cornerstone of Walmart's operations
which emphasizes the transparency and fairness in its interactions with
customers, employees, and partners, fostering trust and loyalty.

 By improve and remain to these core values, Walmart not only aims to be the
preferred shopping destination for customers but also seeks to build strong, lasting
relationships based on trust and respect. The organization goes above and beyond to
ensure it does the right thing, demonstrating a commitment to justice and sincerity
in all its engagements.

7. Customer segments:

• Walmart segmentation, targeting and positioning is the core focus of Walmart

strategic marketing. Segmentation refers to dividing population into groups
according to certain characteristics, whereas targeting is associated with
choosing specific groups identified because of segmentation to sell products.

31
Supply Chain - Demand Forecasting with Machine Learning Group 1

Positioning refers to the selection of the marketing mix the most suitable for
the target customer segment.

• Walmart uses mono-segment type of positioning and accordingly, Walmart

marketing management appeals to single customer segment who place greater
value on the price attribute of products compared to other attributes.

The following table illustrates Walmart segmentation, targeting and positioning:

Table 2: Customer segments

Type of Segmentation Walmart target segment

segmentation criteria
Region Domestic and international
Geographic Density Urban and rural areas
Age Individuals of all age categories
Gender Males and Females
Bachelor Stage young, single people not living
at home
Newly Married Couples young, no children
Full Nest I youngest child under six
Full Nest II youngest child six or over
Full Nest III older married couples with
Life-cycle stage dependent children
Demographic
Empty Nest I older married couples, no children
living with them
Empty Nest II older married couples, retired, no
children living at home
Solitary Survivor I in labor force
Solitary Survivor II retired
Individuals and households with low incomes
Income
and middle class

32
Supply Chain - Demand Forecasting with Machine Learning Group 1

Students, manual workers, floor level

Occupation employees and middle level managers in public
and private sectors
‘Hard core loyal’, i.e. individuals who always
purchase the product / brand in question.
‘Switchers’, i.e. individuals who do not
Degree of loyalty
specifically seek out a particular brand, but
rather purchase the brand available to them at
Behavioral
time of need, or that which was on sale
Benefits sought Cost advantage
Personality Reserved and cost-conscious individuals
non-users, potential users, first-time users,
User status
regular users, or ex-users of a product
Social class Lower class, working and middle class
Resigned, struggler and mainstreamer
Psychographic individuals according to Cross Cultural
Lifestyle
Consumer Characterization by Young &
Rubican

 It is important to specify that although Walmart attempts to target customers

of all ages via its marketing communication messages, particular attention is
paid to young consumers, dubbed as generation X due to the strategic
importance of achieving the loyalty of young consumers for long-term
perspectives.

33
Supply Chain - Demand Forecasting with Machine Learning Group 1

Figure [Link] retailers index with shoppers with age group

As it is illustrated in figure below, Millennials favor Walmart over stores

including Target, Costco, Trader Joe’s, and Whole Foods in North America. We
can notice that along with the promise of cheap prices, Walmart attracts
Millemials via effective integration of social media marketing into its marketing
communications strategy. This strategy is proving to be effective so far.

8. Sales Channel:

8.1. Physical Stores:

• Store Formats: Walmart has different types of stores to meet various shopping
needs. They have Supercenters, which offer a full range of groceries along with
general merchandise. Then there are Discount Stores, which focus on providing
lower-priced goods, and Neighborhood Markets, which are smaller grocery-focused
stores. This variety helps Walmart cater to different customers and their preferences.

• Store Experiences: When you walk into a Walmart, you’ll notice that they
prioritize a friendly shopping experience. The stores have wide aisles, clear signs,
and well-organized product displays. This layout is designed to make it easy for
customers to browse and find what they need, often leading to those impulse buys!

8.2. E-commerce:

• Website and App: Walmart’s website and mobile app are user-friendly, making it
simple to search for products, read customer reviews, and get personalized
recommendations. The app even lets you scan barcodes to check prices and create
shopping lists, which is super handy.

• Home Delivery: For those busy days, Walmart offers home delivery services,
bringing groceries and other products right to your door. This is especially great for
families or anyone who might not have time to go shopping.

34
Supply Chain - Demand Forecasting with Machine Learning Group 1

• Same-day Delivery: In some areas, Walmart even provides same-day delivery for
groceries and other items, putting them in direct competition with services like
Amazon Fresh.

8.3. Marketplace:

• Third-parties seller: Walmart’s online marketplace allows other sellers to list their
products alongside Walmart’s own inventory. This means a wider selection for
customers and helps Walmart compete with bigger online retailers.

• Quality Control: Walmart keeps a close eye on the quality of products sold by
third-party sellers to ensure that customers have a good shopping experience and
that the Walmart brand remains strong.

8.4. Mobile Sales:

• Walmart Mobile Application: The Walmart app isn’t just for shopping; it also
helps you track your orders, find digital coupons, and use Walmart Pay for secure
transactions. It makes the whole shopping process smoother.

• Picking up services: If you prefer not to go inside the store, Walmart offers curbside
pickup. You can place your order online and have it brought right to your car, which
is especially convenient and contactless.

8.5. Membership Program:

• Walmart+: This membership program gives you perks like free shipping on online
orders, discounts on fuel at Walmart gas stations, and early access to special
promotions. It’s designed to keep customers loyal and compete with services like
Amazon Prime.

• Membership’s promotion: Members of Walmart+ can enjoy exclusive deals and

promotions, encouraging them to shop more often and spend a bit more each time.

8.6. Direct Sales:

• Online direct sale: When you shop on Walmart’s website; you can buy a wide range
of products directly from them. This straightforward approach makes it easy and
convenient for customers to find what they need without dealing with third parties.

8.7. Omnichanel:

• Integration: Walmart’s strategy is all about making sure customers can easily
switch between shopping online and in-store. Whether you want to order something
online and pick it up at the store or return an online purchase in person, Walmart
aims to make the experience as smooth as possible.

35
Supply Chain - Demand Forecasting with Machine Learning Group 1

 In summary, Walmart uses a mix of physical stores and online options to meet the
needs of its customers

IV. MODELING:
1. Decision Tree:
A Decision Tree10 is a non-parametric supervised learning model used for
classification and regression. For this sales prediction model, it works by recursively
splitting the data into subsets to predict a continuous output – weekly sales – based on
features such as store type, holiday flag, temperature, fuel price, CPI, and
unemployment data. For instance, temperature could impact weekly sales of seasonal
items, while the unemployment rate might affect overall purchasing behavior in
different regions. Decision nodes could split based on thresholds of temperature or
employment figures to form subsets that align with sales patterns. The tree structure is
made up of nodes:
- Root Node: Contains all training data.
- Decision Nodes: Points where data is split based on features like store.
- Leaf Nodes: Contain the final predicted sales values.
The model selects splits based on metrics like Gini Impurity or Information Gain
(entropy-based) to reduce uncertainty and increase purity at each step. Decision Trees
are intuitive and easy to visualize, but they can be prone to overfitting, especially with
deep trees. Techniques like pruning and ensemble methods (e.g., Random Forests) are
often used to improve their generalization ability.

Figure 9: Decision tree model

• Strengths:

10
Scikit-learn. (n.d.). [Link]. Scikit-learn

36
 Supply Chain - Demand Forecasting with Machine Learning Group 1

- Non-linear relationships: Can capture complex, non-linear patterns in data.

- Interpretable: Decision trees are easy to visualize and understand.
• Weaknesses:
- Overfitting: A deep decision tree trained on historical sales data might memorize
anomalies like temperature or unexpected events, which would impact its
performance on new data.
- Sensitive to noise: Noise in the data can lead to suboptimal decision boundaries.

2. Support Vector Machine (SVM):

The Support Vector Machine (SVM11) model, while commonly associated with
classification, can also be adapted for regression tasks, known as Support Vector
Regression (SVR). The goal of SVR is to find a hyperplane that fits the data points
while keeping prediction errors within a specified tolerance (epsilon). The model
maximizes the margin around this hyperplane, focusing only on data points outside the
margin to refine the prediction.

SVR12 utilizes kernel functions, such as polynomial and radial basis function
(RBF) kernels, to map the data into higher-dimensional spaces, enabling it to capture
complex, nonlinear relationships. While SVR excels in high-dimensional data and
generalizes well, it can be computationally intensive for large datasets, and selecting
the appropriate kernel function is key to its success.

Figure 10: Support Vector Machine (SVM) model

11
Scikit-learn. (n.d.). Support vector machines. Scikit-learn.
12
Scikit-learn. (n.d.). Support vector machine regression example. Scikit-learn

37
Supply Chain - Demand Forecasting with Machine Learning Group 1

• Strengths:
- Effective for high-dimensional data: Can handle complex datasets with many
features.
- Good generalization performance: Often performs well on unseen data.
• Weaknesses:
- Can be computationally expensive, especially for large datasets.
- Choice of kernel function is crucial: The choice of kernel can significantly
impact performance.

3. Random Forest:

Random Forest13 are a popular supervised machine learning algorithm that can
handle both regression and classification tasks. They are particularly well-suited for
business applications like sales forecasting, where multiple features (e.g., fuel price,
unemployment, temperature, etc.) interactly impact the target outcome. Random
forests operate as an ensemble of decision trees, where each tree is trained on a
random subset of data and features, providing a combined prediction that improves
accuracy and reduces overfitting.

• Strengths:

- Handles complex and high-dimensional data: Suitable for datasets with

many influencing features, as in Walmart's weekly sales.

- Reduces overfitting risk: Random Forest’s ensemble nature provides

robustness against overfitting, crucial for demand forecasting.

- Good generalization performance: Typically performs well on unseen data,

offering reliable demand predictions.

- Provides feature importance insights: Helps identify key factors driving

sales, aiding in strategic planning.

- Scalability: Works efficiently with large datasets and can be parallelized for
faster processing.

• Weaknesses:

- Computationally intensive: Building multiple decision trees can be time-

consuming, especially for large datasets.
- Interpretability: Results are less interpretable compared to simpler models,
making it harder to explain predictions.
- Requires parameter tuning: Model performance can depend on tuning
parameters like the number of trees and max depth.

13
Scikit-learn. (n.d.). [Link]. Scikit-learn.

38
Supply Chain - Demand Forecasting with Machine Learning Group 1

Figure 11: Random Forest model

V. EVALUATION METRICS:

1. R-squared 14(R²):
R², or the coefficient of determination, measures the proportion of the variance
in the dependent variable that is predictable from the independent variables. It indicates
how well the model’s predictions match the actual data.
• Formula:

2
∑𝑛𝑖=1(𝑦𝑖 − 𝑦̂𝑖 )2
𝑅 =1− 𝑛
∑𝑖=1(𝑦𝑖 − 𝑦̅𝑖 )2
Where:
𝑦𝑖 = actual values,
𝑦̂𝑖 = predicted values,
𝑦̅𝑖 = mean of actual values.

2. Mean Absolute Error (MAE):

MAE15 measures the average magnitude of the error in a set of predictions, without
considering their direction. It is the average of the absolute differences between the
predicted and actual values.

• Formula:

14
Newcastle University. (n.d.). Coefficient of determination: R-squared. Newcastle University.
15
Statistics How To. (n.d.). Absolute error. Statistics How To

39
Supply Chain - Demand Forecasting with Machine Learning Group 1

𝑛
1
𝑀𝐴𝐸 = ∑|𝑦𝑖 − 𝑦̂𝑖 |
𝑛
𝑖=1

3. Mean Squared Error (MSE):

MSE16 is the average of the squares of the errors – that is, the average squared
difference between the actual and predicted values. It emphasizes larger errors due to
the squaring effect.

• Formula:
𝑛
1
𝑀𝑆𝐸 = ∑(𝑦𝑖 − 𝑦̂𝑖 )2
𝑛
𝑖=1

4. Root Mean Squared Error (RMSE):

RMSE17 is the square root of the mean squared error and provides a measure of how
spread out the residuals (errors) are. It is in the same unit as the target variable, making
interpretation straightforward.

• Formula:
𝑛
1
𝑅𝑀𝑆𝐸 = √𝑀𝑆𝐸 = √ ∑(𝑦𝑖 − 𝑦̂𝑖 )2
𝑛
𝑖=1

VI. RESULT:
• Result for exogenous (external) variables:

Table 3: Result for exogenous(external) variables

Model Decision Tree Support Vector Random Forest

Regression
R-squared -0.1713 0.1593
MAE 393503.38 460281.09 356869.1
MSE 365139009951.1 317778417925.2 262070270683.58
RMSE 604267.33 563718.38 511927.99

Models trained on only exogenous (external) variables show the weakest

performance, with low or negative R-squared values and high error metrics (MAE,

16
Frost, J. (n.d.). Mean squared error (MSE): Definition and interpretation. Statistics By Jim
17
Frost, J. (n.d.). Root mean square error (RMSE): Definition and interpretation. Statistics By Jim

40
Supply Chain - Demand Forecasting with Machine Learning Group 1

MSE, and RMSE). This suggests that external factors alone are insufficient for
capturing the underlying patterns in the data.

Both Random Forest and Support Vector Regression show similar results with a
relatively low R-squared, indicating that the models have limited ability to explain
the variation in the data for exogenous variables. Random Forest outperforms SVR
in terms of MAE and RMSE, suggesting that it makes more accurate predictions.

• Result for internal variables:

Table 4: Result for internal variables

Model Decision Tree Support Vector Random Forest

Regression
R-squared 0.931 0.9309
MAE 85116.12 460200.34 85046.30
MSE 21523971456.99 317644726372.7 21515301780.70
RMSE 146710.50 563599.79 146680.95

The performance significantly improves when models are trained on internal

variables. High R-squared values show that these variables better explain the
variance in the target variable.

All models show high R-squared values (around 0.93), indicating excellent
predictive performance for internal variables. Decision Tree and Random Forest
perform similarly, with close MAE and MSE values, and low RMSE, pointing to
accurate predictions. However, Random Forest shows slightly better performance
in terms of MAE and RMSE.
• Result for all variables:

Table 5: Result for all variables

Model Decision Tree Support Vector Random Forest

Regression
R-squared 0.8884 0.9379
MAE 94902.85 460277.84 74012.76
MSE 34779809608.1 317771371251.6 19353089522.17
RMSE 186493.45 563712.13 139115.38

41
Supply Chain - Demand Forecasting with Machine Learning Group 1

The best performance is achieved when both internal and external variables are
used together. When using all variables, Random Forest indicates the best results
with the highest R-squared value (0.9379) and the lowest MAE, MSE, and RMSE
among all models. This suggests that Random Forest is the most capable model for
combining information from both sets of variables and making the most exact
predictions. While the Decision Tree also works well, it does not achieve the
performance of the Random Forest.

In conclusion, the results demonstrate that internal variables play a more crucial
role in accurate model predictions than external ones. Combining both internal and
external variables deliver the most reliable outcomes, enhancing model performance
across all metrics. In particular, the Random Forest shows superior performance in
terms of prediction accuracy across all three groups of variables, especially when
using all variables. It also minimizes error better than the other models. While
Decision Tree and SVR have certain limitations, particularly when handling
exogenous variables, but they still solve well with internal variables. However,
Random Forest clearly outperforms them in overall accuracy and error reduction.

VII. CONCLUSION AND SUGGESTION FOR FUTURE

DIRECTIONS:
Through running model machine learning to predict demand of customer base
on dataset weekly sale of Walmart. With the supply chain research above, we can apply
model to have some suggestion to upgrade supply chain of Walmart progress general,
the startup or smaller business in specific can reference and apply to their supply chain
to improve the process.

As we saw, the model after run dataset have external value vs internal value have
large distance with each other, so we can see that the external will affect to the revenue
of that store or business.

So, we can use machine learning to run model, predict the demand of customer
base on the history data to improve supply chain such as:

1. Assess the Influence of External Factors:

Conclusion: If the forecast with exogenous variables (such as weather, holidays,

promotions, economic conditions) shows a notable improvement in accuracy over the
forecast without them, it implies these external factors play a crucial role in influencing
demand.

Supply Chain Suggestion: Regularly monitor and integrate relevant external data into
the forecasting process. This can help Walmart proactively manage stock levels,
optimize distribution schedules, and reduce the risk of overstock or stockouts. For
example, anticipate increased demand for specific products before holidays or during
expected weather changes, and adjust inventory accordingly.

42
Supply Chain - Demand Forecasting with Machine Learning Group 1

2. Develop Flexible Inventory Strategies:

Conclusion: The discrepancy between forecasts indicates that a static inventory

approach may not suffice, especially for products heavily influenced by external
conditions.

Supply Chain Suggestion: Implement flexible inventory policies based on demand

forecasts that consider exogenous factors. Introduce agile restocking processes to meet
fluctuating demands quickly. This may include dynamic safety stock levels that adjust
according to real-time external factors like local events or weather conditions.

3. Refine Procurement and Supplier Coordination:

Conclusion: Exogenous variables can lead to spikes in demand that standard forecasts
miss. This indicates a need for quick-response capabilities in procurement.

Supply Chain Suggestion: Strengthen collaboration with suppliers to prepare for

sudden increases or decreases in demand. Consider adopting just-in-time (JIT)
principles with reliable suppliers who can respond quickly to forecast adjustments.
Additionally, establish contingency plans for alternative suppliers to cover unexpected
demand.

4. Implement Region-Specific Strategies:

Conclusion: External factors often vary by region (e.g., weather events, local holidays,
regional economic changes), suggesting that a one-size-fits-all approach may not be
effective.

Supply Chain Suggestion: Use regional forecasts with exogenous variables to tailor
inventory and distribution strategies for each location. For example, adjust stock levels
in stores based on regional forecasts of weather or events, ensuring that demand is met
locally without excessive transportation costs.

5. Enhance Demand Forecasting Models:

Conclusion: The difference in forecast accuracy indicates that exogenous factors are
essential to understanding real-world demand patterns. Relying solely on historical
sales may lead to inaccurate predictions.

Supply Chain Suggestion: Invest in demand forecasting models that allow for
continuous integration of exogenous data, like machine learning models that can
dynamically adjust for seasonal or sudden external impacts. Additionally, maintain data
partnerships or external data sources to update models regularly with relevant external
data.

6. Adjust Pricing and Promotion Strategies:

Conclusion: The forecast difference highlights that external events (like economic
changes or competitive promotions) can significantly influence demand elasticity.

43
Supply Chain - Demand Forecasting with Machine Learning Group 1

Supply Chain Suggestion: Develop dynamic pricing and promotional strategies that
respond to anticipated demand changes due to exogenous factors. For instance,
implement region-specific promotions based on economic conditions, or offer discounts
during adverse weather to stimulate demand.

 Conclusion:

The observed forecast discrepancies confirm that incorporating exogenous variables

leads to more accurate and actionable demand forecasts. Walmart’s supply chain can
benefit from a flexible, responsive approach that leverages these external insights. By
embedding exogenous factor analysis into routine planning, Walmart can stay agile,
minimizing costs while maximizing service levels and customer satisfaction.

These insights help Walmart proactively manage stock and distribution, mitigating risks
associated with sudden demand changes due to external influences. Let me know if
you’d like further details on any of these points!

44
Supply Chain - Demand Forecasting with Machine Learning Group 1

REFERENCES

Index Citation
H, M. Y. (2021, December 26). Walmart dataset. Kaggle.
1
[Link]
Aldahmani, E., Alzubi, A., & Iyiola, K. (2024). Demand Forecasting
in Supply Chain Using Uni-Regression Deep Approximate
2
Forecasting Model. Applied Sciences, 14(18), 8110.
[Link]
Feizabadi, J. (2020). Machine learning demand forecasting and
3 supply chain performance. International Journal of Logistics
Research and Applications, 25(2), 1–24.
Carbonneau, R., Vahidov, R., & Laframboise, K. (2007). Machine
4 Learning-Based Demand Forecasting in Supply Chains.
International Journal of Intelligent Information Technologies, 3(4),
40–57. [Link]
GlobalData. (2022). Company Profile. GlobalData.
5
[Link]
Rickerby, M. (2020, January 4). Walmart supply chain: Winning at
6 inventory logistics. Extensiv.
[Link]
Choo, J. (2024, January 5). Walmart business strategy: A
comprehensive analysis. THE STRATEGY JOURNEY.
7
[Link]
comprehensive-analysis/#0-walmart-business-strategy-
Lövenich, A. (2024, September 23). The amazing supply chain of
8 Walmart. All Things Supply Chain - Supply Chain Trends, Best
Practices, News and Much More!; All Things Supply Chain.

45
Supply Chain - Demand Forecasting with Machine Learning Group 1

[Link]
of-walmart/
Thi Thu Ha, N. (2017). Wal-Mart’s Successfully Integrated
Supply Chain and the Necessity of Establishing the Triple-A Supply
9
Chain in the 21st Century. ResearchGate.
[Link]
Scikit-learn. (n.d.). [Link]. Scikit-
10
learn
11 Scikit-learn. (n.d.). Support vector machines. Scikit-learn.
Scikit-learn. (n.d.). Support vector machine regression example.
12
Scikit-learn
Scikit-learn. (n.d.). [Link].
13
Scikit-learn.
Newcastle University. (n.d.). Coefficient of determination: R-
14
squared. Newcastle University.
15 Statistics How To. (n.d.). Absolute error. Statistics How To
Frost, J. (n.d.). Mean squared error (MSE): Definition and
16
interpretation. Statistics By Jim
Frost, J. (n.d.). Root mean square error (RMSE): Definition and
17
interpretation. Statistics By Jim

46