From Decision Models to Prediction Models: Enhancing Supplier Selection with Machine Learning and Deep Learning

This study extends the traditional Multi-Criteria Decision-Making (MCDM) framework for supplier selection in circular supply chains by integrating predictive machine learning (ML) and deep learning (DL) models. Conceptually, it bridges the gap between interpretable, theory-driven decision models and powerful, data-driven prediction, proposing a hybrid intelligence framework where MCDM provides the foundational "expert judgment" for ML to learn and automate. Building on a foundational MCDM analysis that ranked 100 raw material suppliers for a Nigerian beverage firm using Weighted Sum (WSM), Weighted Product (WPM), and TOPSIS methods, this study treats the derived composite MCDM score as a target variable for predictive modeling. We implemented and compared several ML models, Linear Regression, Random Forest, XGBoost, and Support Vector Regression (SVR), alongside a Deep Neural Network (DNN) to predict supplier performance. The results revealed that a simple Linear Regression model achieved the highest predictive accuracy (MAE = 0.0103), outperforming more complex ensemble and deep learning models. Feature importance analysis from the Random Forest model confirmed the critical role of Sustainability Efforts and Financial Stability, validating the pre-defined weights in the original MCDM framework. The core conceptual contribution lies in demonstrating a viable pathway "from decision to prediction," establishing a novel, automated pipeline for supplier scoring that learns from expert-derived MCDM rankings. This demonstrates that robust predictive analytics can effectively streamline and enhance supplier evaluation processes in circular supply chains, with simpler models often providing superior performance on structured operational data.