Enhancing Retail Sales Forecasting Using Exponential Factorization Machines: A Data-Driven Approach to Demand Prediction

Abstract

Accurate sales forecasting is essential for retailers to streamline inventory management, minimize losses, and maximize profitability. This paper introduces a novel Exponential Factorization Machine (EFM) model designed specifically for predicting retail sales, particularly for new stock-keeping units (SKUs) with extended lead times and short life cycles. Unlike conventional forecasting techniques, EFM leverages product attributes, marketing dynamics, and store-level interactions to enhance predictive performance. The model integrates percentage error minimization (PES) and adaptive batch gradient descent (ABGD) to improve accuracy while reducing risks associated with overstocking and stockouts. Using real-world data from a footwear retailer in Singapore, the results demonstrate that EFM surpasses existing models in terms of Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). This study presents a data-driven forecasting approach, equipping retailers with actionable insights for better decision-making in rapidly evolving market environments.