APPLICATION OF MULTI-CRITERIA OPTIMIZATION FOR FEATURE SELECTION IN MACHINE LEARNING BASED RISK CLASSIFICATION SUPPORTING SUPPLY CHAIN MANAGEMENT

Purpose:The aim of this research is to develop a multi-criteria optimization method for supply chains operating under uncertain conditions.The study addresses the impact of supply chain disturbances (drifts) and proposes an AI-driven approach to enhance risk classification and stability.Design/methodology/approach: The research employs multi-criteria optimization techniques, specifically the NSGA-II genetic algorithm, to optimize feature selection in supply chain risk classification.The methodology includes experiments on global and local optimization of supply chain links, analyzing classification accuracy under different levels of drift.Machine learning models, including deep learning (DNN, CNN), Random Forest, and SVM, are used to assess the effectiveness of the proposed method.Findings: (mandatory) The study confirms that multi-criteria optimization improves supply chain stability and enhances the accuracy of drift classification models.The best feature selection strategies contributed to classification accuracy improvements, particularly in deep learning models, where performance gains of up to 2.5% were observed.The results demonstrate that both global and local optimization contribute to maintaining classification quality even under significant drift conditions.Research limitations/implications: While the study provides strong evidence for the effectiveness of multi-criteria optimization, it does not explore real-time adaptation to evolving risks.Future research should focus on integrating reinforcement learning and real-time monitoring for adaptive optimization.Additionally, extreme drift conditions (above 70%) still challenge classification accuracy, suggesting the need for further improvements in feature selection methods.Practical implications: The research has significant implications for supply chain management, enabling enterprises to improve risk detection and enhance resilience against supply chain disruptions.The proposed optimization approach can support automated decisionmaking systems in logistics, reducing operational risks and improving efficiency in dynamic environments.Social implications: By enhancing supply chain stability, the research contributes to economic resilience, reducing the negative impact of supply chain disruptions on businesses and consumers.Improved risk classification methods can support sustainable supply chain strategies, minimizing delays and reducing waste.Originality/value: This study introduces a novel application of multi-criteria optimization in supply chain risk classification, integrating feature selection with machine learning techniques.The findings provide valuable insights for researchers and practitioners in AI-driven supply chain management, offering new strategies for mitigating risk drift.