PREDICTING SUPPLY CHAIN DISCRUPTION IN TURBULENCE ENVIRONMENTS: A MACHINE LERNING-BASED EARLY WARNING FRAMEWORK FOR LIKELIHOOD AND SEVERITY

Abstract

Global supply chains now operate under persistent turbulence driven by trade wars, geopolitical conflict, climate change, and macroeconomic instability. These shocks are increasingly frequent, correlated, and non-stationary, rendering traditional forecasting and static risk registers insufficient. This article develops and tests a predictive-analytics framework that provides an early-warning capability for supply chain disruptions by estimating (i) the probability that a disruption will occur within a forward horizon and (ii) the expected severity of that disruption if it occurs. The proposed approach integrates (a) exogenous turbulence indicators (tariffs, geopolitical volatility, climate hazard probabilities, port congestion, exchange-rate volatility, demand volatility) with (b) operational and financial fragility indicators (lead-time statistics, inventory coverage, supplier financial stress, and dependency concentration). A two-stage learning structure is employed: a classification model forecasts disruption likelihood, and a conditional regression model predicts severity for disrupted cases. Implemented Random Forest and gradient-boosted decision trees (XGBoost-like) as robust baselines and a sequence-based long short-term memory network to capture temporal dynamics and non-linear interactions. To translate predictions into actionable early-warning insights, calibrated alert thresholds using cost-sensitive optimization and generate forward-looking risk trajectories using scenario-based Monte Carlo simulation of turbulence drivers. Across extensive computational experiments, the framework delivers substantially improved discrimination over naive benchmarks and yields interpretable risk signals that support proactive interventions such as inventory buffering, sourcing switches, and logistics rerouting. The study contributes a rigorous, programmatic methodology for disruption prediction beyond conventional forecasting, and it offers a practical blueprint for deploying predictive resilience capabilities in turbulent supply chain environments.