An Explainable ERP-Integrated Predictive Analytics Framework for Manufacturing Defect Risk Reduction

Abstract

In the manufacturing sector, companies are increasingly turning to enterprise resource planning (ERP) to organise supply chain management, production, maintenance, quality control and inventory management, but ERP systems are usually not proactive in managing defect risk. The study suggests an explainable ERP integrated predictive analytics framework to support the reduction of manufacturing defect risks. The framework, based on 3,240 manufacturing defect records and 16 operational variables, maps variables to ERP domains of risk, engineers 7 composite risk indicators, applies three class-imbalance resampling strategies and assesses nine machine-learning models, including a stacking ensemble of gradient-boosted learners. Random Forest had the best F1 (0.969), recall (0.989), and false-negative rate (0.011) of all the classifiers. SHAP-based explainability pinpointed MaintenanceHours, QualityScore, and ProductionVolume as major drivers for defect risk, and turned them into conditional ERP decision rules for procurement, maintenance, quality, and inventory.