LEVERAGING MACHINE LEARNING FOR SUPPLY CHAIN DISRUPTION PREDICTION AND RISK ASSESSMENT IN U.S. INDUSTRIES

Md Abubakkar

Abstract: Supply chain disruptions have intensified across U.S. industries as transportation congestion, logistics imbalances, and macro shocks propagate through freight networks. This volatility exposes a critical weakness in current monitoring systems: most rely on lagging metrics that fail to provide actionable early-warning signals. This study develops a multimodal prediction and risk assessment pipeline that fuses the Global Supply Chain Pressure Index (GSCPI) with Bureau of Transportation Statistics freight indicators, FRED transportation series, and derived anomaly signals. We benchmark temporal and non-temporal models, including LSTM forecasters, XGBoost, Random Forests, and an Isolation Forest anomaly layer, to detect emerging stress conditions and predict short-horizon disruption states. To ensure operational transparency, we integrate global and local SHAP explanations, enabling attribution of predicted stress to underlying freight congestion, modal throughput changes, and macro-logistics perturbations. We extend analysis with scenario-based stress testing by perturbing transportation exposures and recomputing disruption probabilities to evaluate system sensitivity under hypothetical shocks. Results show that fused models consistently outperform single-source baselines, identifying early disruption signatures with higher recall and stronger temporal calibration. SHAP analysis reveals stable driver patterns linking trucking capacity, freight throughput deviations, and logistics bottlenecks to spikes in predicted pressure. Scenario tests demonstrate distinct vulnerability profiles across industries, supporting the construction of continuous risk scores mapped to Low, Medium, and High categories. Taken together, these findings show that interpretable ML can provide forward-looking, industry-aligned risk intelligence that moves supply chain monitoring from reactive assessment toward predictive resilience.