Sales Forecasting for

a Retail Chain
ICT702 –
Business Analytics &
Visualization

Group Name: -

 Rock

Group Members: -

 Drishtant Karki
 Riyyan Uddin Mohammed
 Salahuddin Mohammed
 Abhishek Bhattarai
Introduction

Retail businesses have to forecast sales in the future in an attempt to control inventory, workers,
and prices. Poor forecasting of sales will result in stockouts or excess stock, and that equates to
profits. The project examines 45 stores' sales data considering holiday and discount influences on
demand. The Kaggle data includes sales patterns due to the Super Bowl, Labor Day,
Thanksgiving, and Christmas. Through machine learning and data analytics, we will create a
model that will enable companies to predict sales more precisely. The companies can then make
decisions using this, save waste, and boost profit. The project also enables the students to learn to
analyze big data and make business decisions based on business analytics [1].
Problem Statement

Traders struggle to predict sales with precision, especially during festive seasons and
promotions. If they make too optimistic predictions, they will be left with surplus stock, which
will have storage fees attached and end up as waste. If they predict too low, they will lose sales
and annoy customers. This project will work on historical sales figures of 45 stores and design a
model for forecasting future sales based on these. The data contains sales by various departments
and monitors the effects of promotions as well as notable holidays such as the Super Bowl, Labor
Day, Thanksgiving, and Christmas.

Since holiday weeks are five times more important in terms of rating, it is necessary to
understand these trends. We try to answer significant questions:

 How do discounts affect sales?

 What are the trends in sales patterns?
 Can we be sure of future demand?

A good sales forecasting model will help stores manage stock, plan promotions, and maximize
profitability.

Data Sources & Preprocessing

Data Source

We will use the Online Retail Dataset on Kaggle for this project. It contains sales records from
45 stores and how various departments trend over time. It also contains promotional discount
data applied before major holidays, which will assist us in learning about how sales vary during
such holidays [2].

The data contains:

 Department-by-department sales for different stores.

 Transaction dates to track trends over time.
 Discounted promotions to analyze the impact of discounts.
This data will allow us to examine sales patterns, seasonal trends, and consumer trends.

Data Preprocessing

We will prepare and clean the data before using it:

 Handling missing values to ensure accuracy.

 De-duplication and error removal for improved data quality.
 Coming up with new features like holiday indicators to track seasonal impacts.
 Normalizing data such that all the values are of the same type.

Proper data cleansing will allow us to easily set up the right sales forecasting model so that we
can better identify patterns and forecast sales outcomes.

Analysis Methodology

To predict future sales, we will utilize data analysis and machine learning methods. We have a
four-step process:

1. Exploratory Data Analysis (EDA)

 We will observe sales patterns over time to know what people are purchasing.
 We will also discover how holidays and promotions affect sales.
 We will use charts and graphs to analyse trends [3].
2. Model Selection & Training

We will compare some forecasting models to find the best model: Time Series Models (such as
ARIMA and exponential Smoothing) to find out temporal trends. Machine Learning Algorithms
(Linear Regression, Random Forest, XGBoost) to predict future sales based on past data [4].

3. Model Evaluation

To verify the accuracy of our model's sales prediction, we will utilize:

 Root Mean Squared Error (RMSE) to gauge the accuracy of prediction.

 Mean Absolute Percentage Error (MAPE) to identify the extent of difference between
predictions and actual sales [5].

4. Data Visualization & Insights

We will build interactive dashboards to visualize major findings. The outcomes will enable
retailers to plan stock, correct prices, and optimize sales tactics.

Through this process, we are hoping to build a robust sales forecasting model that enables
companies to make better decisions.

Technical, Ethical, and Operational Considerations

This project entails the analysis of extensive sales data to generate helpful business insights. In
order to get reliable results, the data needs to be collected, processed, and analyzed with the right
methods. One should present findings in clear and effective ways so that businesses can easily
comprehend and utilize them in making decisions.

Ethically, privacy and security of data need to be ensured at every step. Although the dataset is
not based on personal customer data, all data needs to be treated responsibly and there should be
no misinterpretation or bias in the analysis. Ensuring fairness in prediction will enable various
stores to gain equally from the insights.
Operationally, the forecast model must be easy to use, feasible, and scalable to meet varied
business requirements. It must enable retailers to plan their inventory, minimize waste, and
enhance efficiency in general [6].

Expected Outcomes

Through this project, the company can benefit from the availability of a good and trustworthy
model for sales forecasting that promotes better decision-making. The model will skillfully learn
how to predict sales forecasts from past data, thus enabling retailers to carry out activities like
inventory management, loss reduction, and profit maximization. Also considered will be the
effects of festivals and promotions on sales so that companies can prepare promotions
accordingly. In addition to this, reports will be generated, as well as graphical dashboards that
place key insights and centers in front of the retailers. These insights will drive action on the
retailers' pricing, inventory, and advertising campaigns. The project eventually resolves a
pressing need aimed at scaling sales planning and enhancing business performance.
Conclusion

Retail business projects predict scenarios like stock-out and overstock. The project intends to
develop a demand forecasting model which will help the retailers in predicting better on future
demands taking into account past sales history and holiday sales. The project will help businesses
streamline revenue returns by analyzing and visualizing data for stock management. The research
synchronizes with business analytics and decision-making intentions, adding real-world
situations to retailers.

Group Contribution Breakdown

Group Members Contribution Percentage

Drishtant Karki Data preprocessing, EDA, and feature engineering 25%

Riyyan Uddin Problem definition, methodology, and report writing 25%

Mohammed

Salahuddin Mohammed Model evaluation, visualization, and results analysis 25%

Abhishek Bhattarai Business insights, ethical considerations, and 25%

documentation
References

[1] A. Krishna, V. Akhilesh, A. Aich, and C. Hegde, “Sales-Forecasting of Retail Stores Using
Machine Learning Techniques,” in 2018 3rd International Conference on Computational Systems
and Information Technology for Sustainable Solutions (CSITSS), Dec. 2018, pp. 160–166. IEEE.
Available: https://ieeexplore.ieee.org/abstract/document/8768765/.

[2] Retail Data Analytics, Kaggle, Sep. 1, 2017. Available:

https://www.kaggle.com/datasets/manjeetsingh/retaildataset?resource=download.

[3] M. Attobrah, “Exploratory Data Analysis,” in Essential Data Analytics, Data Science, and
AI: A Practical Guide for a Data-Driven World, Berkeley, CA: Apress, 2024, pp. 47–73.
Available: https://link.springer.com/chapter/10.1007/979-8-8688-1070-1_4.

[4] O. G. Atanda et al., “Intelligent Sales Forecasting System Using ARIMA, SARIMA, and
XGBoost Models,” in 2024 International Conference on Science, Engineering and Business for
Driving Sustainable Development Goals (SEB4SDG), Apr. 2024, pp. 1–8. IEEE. Available:
https://ieeexplore.ieee.org/abstract/document/10629780/.

[5] A. S. Temür and Ş. Yıldız, “Comparison of forecasting performance of ARIMA, LSTM, and
HYBRID models for the sales volume budget of a manufacturing enterprise,” Istanbul Business
Research, vol. 50, no. 1, pp. 15–46, 2021. Available:
https://dergipark.org.tr/en/pub/ibr/issue/64325/976588.

[6] A. Magrini, “The Role of Predictive Analytics in Sales Forecasting,” SSRN, 2023. Available:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5006903.