Using MCDM Methods to Optimise Machine Learning Decisions for Supply Chain Delay Prediction: A Stakeholder-centric Approach

Background: This study addresses challenges faced by supply chain stakeholders who lack expert knowledge in making decisions related to Machine Learning. It introduces a novel use of Multi-Criteria Decision-Making as an evaluation mechanism for different classifiers, aiding stakeholders in selecting appropriate Machine Learning models to predict supply chain delays.

Methods: The proposed methodology involves applying classifiers (Decision Tree, Bagging, AdaBoost, Random Forest) and evaluating them using quantitative and qualitative metrics. MCDM methods (TOPSIS, MARCOS, COCOSO, MABAC) rank these Machine Learning models, facilitating accessible decision-making for stakeholders. A pharmaceutical industry case study is employed to validate the approach, utilizing Python for analysis.

Results: The case study results confirm the effectiveness of the proposed approach, combining Multi-Criteria DecisionMaking with Machine Learning in order to facilitate stakeholder decisions on suitable algorithms for predicting supply chain delays. The Random Forest classifier is identified as the most balanced option in the context of the case study, and a clear rationale can be provided in support of or against each option through the comparison of metrics, validating the approach's practical applicability and effectiveness.

Conclusions: The combination of Multi-Criteria Decision-Making with Machine Learning provides a significant advancement in empowering stakeholders in supply chain management, particularly those lacking in-depth Machine Learning expertise. This approach enhances decision-making in model selection and has the potential to improve supply chain efficiency.