Real-Time Intelligent Supply Chain Demand Forecasting and Dynamic Repricing System Using Machine Learning

Abstract

In modern retail environments, effective supply chain management requires accurate demand forecasting and adaptive pricing strategies to handle demand uncertainty and inventory fluctuations. This paper presents a real-time intelligent supply chain demand forecasting and dynamic repricing system based on machine learning techniques. The proposed system leverages the Walmart Sales Forecasting dataset, integrating historical sales data, weather-related attributes, and holiday indicators to predict product demand across multiple store locations and departments. Feature engineering is performed using time-series lag variables and rolling statistics to capture seasonality and temporal dependencies. A Light Gradient Boosting Machine (LightGBM) regression model is employed for demand forecasting due to its efficiency and suitability for structured retail data. The forecasted demand is further utilized in a dynamic repricing module that adjusts prices based on demand–inventory imbalance, demand elasticity, and simulated competitor pricing. To emulate real-world deployment scenarios, a real-time sales event simulation module is implemented, representing a Kafka-compatible streaming architecture. The trained model is exposed through a RESTful API using FastAPI, enabling on-demand predictions. An interactive dashboard provides visual insights into demand trends, pricing recommendations, and inventory risk alerts to support data-driven decision-making. Experimental results demonstrate that the proposed system effectively captures demand patterns and offers practical pricing recommendations, making it a scalable and deployable solution for intelligent retail supply chain optimization.