AI-Powered Forecasting for Supply Chain Resilience: Applications of Logistic Regression, Random Forest, and XGBoost in the U.S. Context

Abstract

This study observed that supply chain resilience is a strong determinant of U.S. competitiveness, i.e., in the face of global swamping, such as pandemics, geopolitical conflicts, and climate-related disruptions. More conventional forecasting models, such as ARIMA and Exponential Smoothing (ETS), continue to see widespread use, but are increasingly applied in contexts that minimize nonlinear and unstable drivers and in information-rich environments. The current work contrasted the predictive capability of three machine learning (ML) models, including Logistic Regression (LR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), in forecasting the U.S. retail supply chains and compared them with classical models. Empirical analysis, which relied on the retail demand and retail logistics operations, tested the model results in terms of accuracy, inventory optimization, service reliability, and the provision of cost-efficiency. Through the experiments, it was established that, on average, XGBoost reduced the forecast error by approximately half, inventory costs by roughly 36%, and fill rates to over 95%, and reduced fuel costs by an average of 14%. Random Forest achieved moderate returns, and in a few cases, Logistic Regression underperformed. Ensemble and boosting-based algorithms are the most strategically valuable for forecasting and have the most significant impact on operational efficiency and sustainability. Their combination can contribute significantly to the U.S. supply chain resilience and to global competitiveness.