Forecasting Retail Sales using Machine Learning Models

Oluwasola Oluwaseun Mustapha () and Dr. Terry Sithole ()

American Journal of Statistics and Actuarial Sciences, 2025, vol. 6, issue 1, 35-67

Abstract: Purpose: This paper’s main objective is to examine common machine learning techniques and also time series analysis for sales forecasting in a bid to get the best fitted technique and give more logical hypotheses for raising future profit margins while obtaining historical in-depth understanding of prior demand utilising business intelligence software’s like Tableau or Microsoft Power BI. The outcomes are laid forth with regards to dependability as well as precision of the various forecasting models that were employed. Materials and Methods: In this project, a sales prediction is carried out on a 5 year store-item sales data for 50 different items in 10 different stores with a dataset obtained from Kaggle. This study focuses on using Machine Learning Methods including the Random Forest, Gradient Boosting Regression (XGBoost), Linear Regression and also the standard time series Autoregressive Integrated Moving Average (ARIMA) method were analysed and contrasted to measure the methods’ effectiveness for prediction of Sales. Findings: This study demonstrates the potential of machine learning algorithms in accurately forecasting sales, which can be extremely valuable for businesses in optimizing their operations, inventory management, and financial planning. By leveraging these predictive models, companies can make data-driven decisions to improve efficiency, reduce costs, and increase profitability. The findings also highlight the importance of selecting the most appropriate algorithm for a given dataset and problem, as well as the need for proper model tuning and validation to ensure reliable results. Furthermore, the study underscores the significance of understanding and interpreting error metrics like RMSE and MAE to effectively evaluate and compare model performance. Unique Contribution to Theory, Practice and Policy: Factors such as Seasonality, Trend, Promotional offers and Randomity have been known to be important factors that affect the outcome of Sales Forecasting which is why the performances of the Mean Absolute Error (MAE), the Absolute error (R2) and the Root Mean Square Error (RMSE) are all compared in the different algorithms used, to help identify the best preferred algorithm to be adopted which turned out to be the XGBoost method.

Keywords: Sales forecasting; Machine learning; Time series analysis; Random Forest; ARIMA; LSTM; XGBoost; Linear regression; RMSE; and MAE. (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
https://ajpojournals.org/journals/index.php/AJSAS/article/view/2679 (application/pdf)
Access to full texts is restricted to American Journal of Statistics and Actuarial Sciences

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:bfy:oajsas:v:6:y:2025:i:1:p:35-67:id:2679

Access Statistics for this article

More articles in American Journal of Statistics and Actuarial Sciences from AJPO Journals Limited
Bibliographic data for series maintained by Chief Editor ().