Supply Chain Optimization With Hybrid Particle Swarm Optimization And Xgboost

Abstract

Forecasting and Distribution/Inventory planning are both critical for supply chain management for optimum service level management and cost reduction. However, the conventional techniques fail to address complex non-linear and dynamic requirements. In this study, we propose a hybrid PSO and XGBoost technique that combines the predictive analysis techniques with optimization to make supply chain management more efficient and cost-effective. For the experiment, data on the historical performance of the supply chain contained in the DataCo Smart Supply Chain data set have been considered. These include sales, inventory, distribution at warehouses, and characteristics of products. Preprocessing includes handling missing values, converting categorical into numerical values, and normalizing the data. Demand forecasting was carried out using the XGBoost model for non-linear relationships and time-series forecasting, while Particle Swarm Optimization (PSO) was utilized for optimal inventory and allocation decisions. The proposed hybrid framework yielded better performance than conventional approaches, with an RMSE value of 9.87, MAE of 7.21, MAPE of 11.3%, service level of 95.8%, and cost of ₹1,102,300, which represents a 12% improvement over using the XGBoost model alone and a 8% improvement over using the PSO model alone. The results demonstrate enhanced forecast accuracy, operational efficiencies, and cost savings. The hybrid approach of PSO-XGBoost demonstrates the ability to combine forecasting models with optimization in the supply chain domain. It provides valuable insights into inventory and logistics management. Future research should explore the potential of extending the proposed framework through deep learning, multi-objective optimization, and real-time adaptive mechanisms.