A Hybrid Machine Learning Approach to Supplier Segmentation and Anomaly Detection for Proactive Circular Supply Chain Risk Management

In the context of circular supply chains (CSCs), which demand a broader evaluation of suppliers beyond cost and quality, this study proposes and demonstrates a hybrid machine learning (ML) framework. The framework synergistically combines unsupervised and supervised learning to move beyond the static scores generated by traditional Multi-Criteria Decision-Making (MCDM) methods. Utilizing a dataset of 100 raw material suppliers for a Nigerian beverage firm, the framework first applies K-Means clustering to discover natural supplier segments, revealing two distinct strategic archetypes, namely Sustainable but Deliberate suppliers and Cost-Effective & Agile suppliers. A Random Forest classifier is then trained to automatically assign suppliers to performance tiers (Low or Mid) based on MCDM-derived rankings, achieving 85% accuracy. Finally, an ensemble anomaly detection system using Isolation Forest and Local Outlier Factor identifies 14 suppliers exhibiting high-risk behavioral patterns. The synthesis of these methods creates an integrated risk-segmentation matrix, enabling stratified management strategies. A critical finding was that 7 out of 14 anomalous suppliers were in the Mid-Tier, highlighting significant risks within otherwise acceptable performers. The study concludes that this hybrid ML approach provides a dynamic, nuanced, and actionable intelligence system for proactive risk management, transforming supplier management in CSCs from a reactive, score-based exercise to a strategic, foresight-driven capability.