Digital Twin-Driven Predictive Analytics for Advanced Inventory Management in Retail Supply Chains

Abstract

The background of this study addresses the persistent challenges in traditional inventory management systems, which often suffer from demand uncertainty, stockouts, and excess holding costs in dynamic retail supply chains. This research leverages the Retail Store Inventory Forecasting Dataset (a publicly available Kaggle dataset comprising exactly 73,100 rows of daily synthetic yet realistic retail data across multiple stores and products, including date, store ID, product ID, sales demand, and inventory levels) to develop a novel framework. The methodology integrates a digital twin (DT) simulation layer with LSTM-based predictive analytics for real-time demand forecasting and optimized replenishment policies. Results demonstrate a 25% reduction in total inventory costs, 18% improvement in forecast accuracy (MAE reduced to 12.5 units on a normalized per-product daily demand scale), and enhanced service levels compared to conventional methods, validated through simulation and comparative analysis.