ABSTRACT

 Effective inventory management is critical for retail businesses aiming to maintain optimal stock levels, minimize costs, and meet fluctuating customer demands. Traditional reorder point (ROP) methods, while simple and widely used, rely on static calculations that fail to adapt to the dynamic nature of modern retail environments. This study proposes a machine learning-based approach to predictive reorder point optimization, leveraging historical sales, stock movement, and product category data from LEO MART SUPERMARKET. The research employs regression and time-series forecasting models, including Linear Regression, Decision Trees, Random Forests, XGBoost, and LightGBM, to dynamically determine optimal reorder points. Feature engineering techniques such as sales velocity, stock turnover rate, and depletion rate are incorporated to improve predictive performance. The proposed system is evaluated using MAE, RMSE, and R 2 metrics, with results as MAE= 4.49, MSE = 45.34, and R2 Score= 0.9776 which are compared against traditional ROP calculations. Results indicate that the machine learning models provide superior accuracy, adaptability to demand fluctuations, and enhanced decisionmaking capabilities, ultimately reducing stockouts and overstocking. This work demonstrates the practical feasibility of integrating machine learning into retail inventory systems, offering a datadriven pathway toward operational efficiency and cost reduction