Journal of Advances in Mathematics and Computer Science

Managing continuous inventory represents a major challenge for commercial enterprises, as it requires ensuring product availability while controlling the costs associated with stockouts and overstocking. Traditional approaches, such as the Wilson model (EOQ), allow the determination of the economic order quantity; however, they remain limited when dealing with dynamic demand fluctuations. This study proposes a hybrid approach that combines the principles of mathematical inventory management with supervised machine learning models, including Random Forest, Support Vector Machine, and Gradient Boosting, integrated within a stacking framework to improve demand forecasting.The experiments are conducted using a dataset extracted from an inventory management system of a large commercial enterprise available on the Kaggle platform. The dataset includes information related to products, sales transactions, orders, dates, and replenishment statuses. The performance of the models is evaluated using statistical metrics such as the coefficient of determination (R²), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE), enabling an objective assessment of forecasting accuracy. The results demonstrate that the combination of classifiers significantly improves prediction accuracy compared with individual models. The developed models are integrated into a responsive web-based business intelligence application that provides inventory managers with real-time visualization of key indicators, including products at risk of stockouts, optimal order quantities, and variations in average stock levels. This decision-support interface facilitates inventory monitoring and enables more proactive supply planning. Overall, the results indicate that the proposed approach improves the reliability of demand forecasts and supports more efficient replenishment decisions while reducing operational costs associated with stock shortages and excess inventory.