Temporal Supplier Risk Prediction and Prioritization

Supplier performance directly affects manufacturing continuity, cost and delivery commitments, yet conventional scorecards are retrospective and may hide gradual deterioration. This paper presents a temporal supplier risk prediction and prioritization framework that uses Month N operational behaviour to estimate a supplier’s Low, Medium or High risk category in Month N+1. The workflow covers supplier-month aggregation, 47 temporal features (lags, rolling statistics, trend and deterioration indicators, risk history and business exposure), engineered target labels, chronological validation and a ninemonth out-of-time (OOT) test. Five classifiers are compared with a persistence baseline on 1,450 supplier-month observations spanning 39 months. HistGradientBoosting achieves the best overall OOT result (54.32% accuracy, 48.93% Macro-F1, 0.7039 ROC-AUC), while Extra Trees reaches 85.51% high-risk recall, showing a trade-off between overall performance and high-risk screening. Predictions are combined with delivery, quality, exposure and deterioration signals into a transparent priority score and rule-based intervention recommendations delivered through a Power BI dashboard. Because the labels are engineered rather than independently observed, the system is positioned as an early-warning aid for human review