Enhancing Grocery Demand Forecasting with Machine Learning and External Factors

Abstract Accurate demand forecasting is vital for grocery retailers to reduce wastage, avoid revenue loss, and meet customer needs. This study applies Machine Learning (ML) techniques to predict grocery demand using historical sales and external factors such as promotions, holidays, and prices. Four models—Linear Regression, Random Forest, XGBoost, and LightGBM—were evaluated with metrics including MAE, RMSE, MAPE, and R². Results show that ensemble models significantly outperform Linear Regression, with LightGBM achieving the best accuracy, while Random Forest provided strong interpretability. Feature analysis highlights promotions and lagged sales as key predictors. The findings demonstrate the potential of ML-based forecasting frameworks to improve inventory management and operational efficiency in the grocery retail sector. Keywords: Grocery Demand Forecasting, Machine Learning, Linear Regression, Random Forest, XGBoost, LightGBM, Ensemble Models, Retail Analytics, Inventory Optimization