Predictive Analytics in Supply Chain Management: Enhancing Demand Forecasting and Operational Resilience under Uncertainty

Abstract

Supply chains have been pressured, in recent years, by a sequence of disruptive events whose timing and magnitude were difficult to anticipate. From the prolonged shockwaves of the COVID-19 pandemic through the Suez Canal obstruction, the Russia–Ukraine conflict, and the persistent volatility in commodity markets, firms have repeatedly discovered that classical forecasting techniques, calibrated on stationary or near-stationary demand patterns, fail precisely when reliable guidance is most needed. This paper examines how predictive analytics, broadly understood to include statistical learning, machine learning, and deep learning approaches, can be deployed to strengthen demand forecasting and operational resilience under uncertainty. Drawing on a mixed-methods design that combines a structured review of 142 peer-reviewed studies published between 2015 and 2025 with an empirical comparison conducted on a multi-echelon retail dataset, we evaluate seven forecasting models: ARIMA, exponential smoothing state space (ETS), Prophet, XGBoost, Long Short-Term Memory (LSTM) networks, Temporal Fusion Transformers (TFT), and a hybrid ensemble. The empirical results indicate that the TFT and the hybrid ensemble outperform classical benchmarks by 18.6% and 22.4%, respectively, in mean absolute percentage error during volatile demand windows, while also yielding tangible improvements in fill rate, inventory turnover, and bullwhip dampening. Beyond model accuracy, we argue that predictive analytics contribute to resilience only when embedded within an organizational architecture that links forecasts to flexible sourcing, dynamic safety stock policies, and human judgment. We propose an integrative framework, the Predict–Sense–Adapt (PSA) loop, and discuss its managerial, theoretical, and policy implications. The study contributes to the literature on data-driven operations management by clarifying when, why, and under what conditions advanced forecasting techniques translate into resilience gains, and where their limitations remain consequential.