Explainable Machine Learning-Driven Supplier Risk Prediction Using ERP Procurement Data and Power BI Decision Dashboards: Evidence from Manufacturing Supply Chains

Abstract

Manufacturing supply chains face supplier risk from delivery delays, quality failures, invoice mismatches, lead-time instability, purchase dependency and disruption exposure. Traditional supplier scorecards are usually descriptive, use fixed weights and often identify supplier risk only after operational disruption becomes visible. This study develops and evaluates an explainable machine-learning framework for supplier risk prediction using real anonymized ERP procurement data from one manufacturing firm. The dataset contains the full active supplier population retained after cleaning: 100 active suppliers observed from January 2023 to December 2024, including delivery, quality, invoice, fulfilment, dependency and disruption-related indicators. A stratified 80:20 train-test split was used, with five-fold cross-validation and 1,000-iteration bootstrap resampling to strengthen robustness under the modest high-risk class size. The study compares a traditional supplier scorecard with Logistic Regression, Random Forest and XGBoost. It achieved the strongest test performance, with 89.7% accuracy, 87.9% precision, 86.5% recall, 87.2% F1-score and 0.928 ROC-AUC. McNemar testing confirmed that XGBoost significantly improved classification over the traditional scorecard. SHAP was used to explain global and supplier-level risk drivers, while Power BI translated model outputs into supplier risk scores, alerts, trend views and mitigation priorities. Moderation analysis further showed that supplier dependency significantly amplified the operational impact of predicted supplier risk (beta = 0.214, p = 0.018). An expert panel review with eight procurement and supply-chain professionals further supported the interpretability and decision-support utility of the framework. The study contributes a Q1-style decision-support framework that links predictive accuracy, explainability, class-imbalance-aware validation and dashboard-based managerial action in manufacturing procurement.