DESIGN AND DEVELOPMENT OF AN AI-BASED RETAIL STORE MANAGEMENT INFORMATION SYSTEM FOR DEMAND FORECASTING AND ORDER RECOMMENDATION

Abstract

Uncertainty in demand within retail store inventory management can lead to both stockouts and overstocks. This study develops a retail store management information system that integrates transaction and stock management with demand forecasting and order recommendations. The system was developed using the Waterfall method, while the prediction module utilized the CRISP-DM framework to compare XGBoost and Holt-Winters, subsequently selecting the model with the lowest MAE for each product. Safety stock and reorder points were calculated from historical sales data, whereas the suggested order quantities were derived from 14-day forecasts. The synthetic dataset for the period of September 2025–August 2026 includes 20 products, 5 suppliers, and 10,886 transactions. The average MAPE was 36.07% for XGBoost and 35.51% for Holt-Winters, indicating that their performances are relatively comparable. Black box testing on 35 scenarios showed that all functions operated as designed. The evaluation results only reflect performance in a simulated scenario, so testing on real transaction data is needed in future research.