Optimizing E-commerce inventory management using a machine-learning approach

E-commerce inventory management faces persistent challenges such as overstock and stockouts caused by unpredictable demand. Traditional inventory systems often fail to process large-scale transactional data efficiently, limiting accurate forecasting and decision-making. To address this issue, this study proposes an integrated machine-learning framework that combines predictive analytics and customer segmentation to improve forecasting precision and inventory control. Three machine learning models LSTM, XGBoost, and Random Forest were compared for demand forecasting. Among them, LSTM achieved the lowest RMSE (0.799), indicating superior predictive performance for time-dependent data. In addition, clustering algorithms, including DBSCAN and K-means, were applied to segment customers based on purchasing behaviour, with DBSCAN achieving a Silhouette Score of 0.9708, suggesting well-separated clusters. The results were visualised to generate actionable insights, enabling data-driven decisions. The findings provide an added approach for e-commerce businesses by linking sales forecasting and customer clustering to more efficient inventory allocation.