A Hybrid XGBoost-LSTM Framework for Supply Chain Demand Forecasting: Empirical Evidence from Retail Multi-Store Data

Abstract

Supply chain demand forecasting is a core component of modern enterprise operations management, directly impacting inventory optimization, production planning, and customer satisfaction. Traditional statistical methods such as Autoregressive Integrated Moving Average (ARIMA) models face limitations when handling complex nonlinear patterns and multidimensional features. While single machine learning models enhance prediction accuracy, they struggle to simultaneously capture structured features and temporal dependencies.This study proposes an innovative hybrid forecasting framework integrating Extreme Gradient Boosting (XGBoost) with Long Short-Term Memory (LSTM) networks. XGBoost excels at learning high-dimensional structured features, while LSTM specializes in modeling temporal dependency patterns. An adaptive weight fusion mechanism enables complementary strengths between the two models.Empirical analysis using the Kaggle open-source retail dataset (encompassing 10 stores, 50 products, and 913,000 transaction records) demonstrates that the hybrid framework significantly outperforms both single models and traditional methods across multiple metrics, including Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE).The findings provide supply chain managers with actionable forecasting tools, offering significant theoretical and practical value for enhancing prediction accuracy and operational efficiency.