A COMPARATIVE MACHINE LEARNING FRAMEWORK FOR RETAIL SALES FORECASTING USING ENSEMBLE TECHNIQUE

Abstract

Accurate forecasting of retail sales is essential for effective inventory management, supply chain coordination, and revenue optimization. However, modeling retail data is challenging due to inherent non-linearities and complex feature interactions. This study proposes a structured machine learning framework for product-level sales prediction using the BigMart dataset, integrating data preprocessing, feature engineering, and comparative model evaluation. Five regression models: Linear Regression, Support Vector Regression (SVR), Decision Tree, Random Forest and Extreme Gradient Boosting (XGBoost) are implemented and assessed using RMSE and R² metrics. Experimental results reveal that ensemble methods significantly outperform individual models, with XGBoost achieving the best performance (R² ≈ 0.924) and the lowest prediction error. The study demonstrates that advanced ensemble techniques effectively capture intricate relationships among product attributes, pricing, and outlet characteristics. The proposed framework provides a scalable solution for retail analytics and supports the development of reliable decision-support systems