An Interpretable Hybrid GRA-BWM Framework with Adaptive Weighting for Quality Cost Enabler Assessment in Sustainable Manufacturing

Abstract

This study propose an Interpretable Ensemble GRA-BWM (IIE-GRA-BWM) framework for prioritizing quality cost
enablers in sustainable manufacturing systems, addressing the limitations of conventional multi-criteria decision-making (MCDM) methods in balancing robustness and interpretability. The proposed method integrates an ensemble of Grey Relational Analysis (GRA) and Best-Worst Method (BWM) instances, dynamically weighted via gradient-boosted decision stumps, with a rule-based refinement module to generate transparent and auditable prioritization outcomes. The ensemble layer mitigates bias and variance by aggregating diverse GRA-BWM outputs trained on perturbed data subsets, while the dynamic weighting mechanism adapts to feature importance shifts through iterative optimization. Furthermore, SHAPbased rule extraction translates complex ensemble decisions into human-readable rules, enabling stakeholders to trace the rationale behind prioritization rankings. The framework uniquely combines the structured foundation of GRA-BWM with the adaptability of ensemble learning, hence overcoming the rigidity of traditional MCDM approaches in dynamic manufacturing environments. Experimental validation demonstrates its superiority in handling categorical and non-linear interactions, a critical requirement for real-world quality cost management. Moreover, the integration of Cat Boost and Kernels HAP ensures computational efficiency and scalability, making it suitable for large-scale industrial deployments. The method’s interpretability meets regulatory and stakeholder demands for transparency, while its modular design allows seamless integration with existing manufacturing data pipelines. This work contributes a novel, hybrid MCDM-ensemble architecture
that advances both theoretical and practical aspects of quality cost prioritization in sustainable production systems.