Supplier selection is a key element of trading and supply chain management, with direct implications for cost, operational efficiency, and long-term performance. The evaluation process involves multiple and often conflicting criteria, and traditional multi-criteria decision-making (MCDM) methods, such as the Measurement of Alternatives and Ranking according to Compromise Solution (MARCOS), rely heavily on subjective expert judgments when determining criteria weights. This study introduces an integrated approach that combines machine learning (ML) algorithms—Random Forest, Decision Tree, Linear Regression, and Gradient Boosting-with the MARCOS method to generate objective criteria weights for supplier evaluation. The approach is applied to a real-world case study using 11,846 procurement records from one of the business wings of Akij Resource, Bangladesh. Feature-importance scores derived from ML models guide the weighting process, with Random Forest selected as the most reliable model based on interpretability and performance metrics, including a mean absolute error of 5.546 and an R 2 score of 0.588. The resulting ML-MARCOS rankings are compared with AHP and TOPSIS, showing improved transparency and ranking stability. Sensitivity analysis indicates that the method maintains consistent performance under variations in criteria weights. By integrating data-driven insights with domain expertise, the ML-MARCOS method offers a robust and objective tool for supplier selection, enabling decision-makers to optimize supply chain operations more effectively while preserving interpretability and expert involvement.
Biswas et al. (Sun,) studied this question.