SD-TS-RF (Sales Data-Time Series-Random Forest) Hybrid Machine Learning Model for Enhanced Next-Day Sales Prediction in Supermarkets

Accurate sales prediction is critical for optimizing inventory management and resource allocation in supermarkets. This study presents an innovative hybrid model, SD-TS-RF, designed to improve the precision of next-day sales forecasting. The model combines Time Series (TS) analysis, Support Vector Regression (SVR)-based Feature Selection (FS), and Random Forest (RF) to deliver superior predictive performance. By analyzing historical sales data alongside contextual factors such as holidays, promotions, and weather conditions, the FS component isolates feature that have the most significant impact on sales outcomes. Empirical evaluations demonstrate the SD-TS-RF model's robustness, achieving a remarkably low Mean Absolute Percentage Error (MAPE) of 2.5%. When benchmarked against conventional Time Series methods, the SD-TS-RF model consistently outperforms, thanks to its strategic feature selection and ensemble learning approach. Beyond enhancing sales forecast accuracy, this model offers a flexible and scalable framework applicable across diverse forecasting scenarios in the retail sector. The findings underscore the SD-TS-RF model's potential as a transformative tool in predictive analytics, enabling smarter, data-driven decision-making in retail demand forecasting.