A multi-objective variable selection framework for enhanced Mahalanobis–Taguchi system-based multivariate process control

Purpose This study proposes a robust multi-objective optimization (MOO)-based variable selection framework integrated with the Mahalanobis–Taguchi System (MTS) for multivariate process quality control (MPQC) under non-normal conditions and varying response dimensionality. The framework addresses key limitations of existing MTS-based approaches that rely on single-objective or weighted-sum optimization with fixed thresholds, which often yield unstable and suboptimal monitoring performance. Design/methodology/approach The proposed MOO–MTS framework employs the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to simultaneously optimize multiple conflicting objectives, including monitoring performance, dimensionality reduction, signal separation, and variable relevance. A repeated nested k-fold cross-validation (CV) scheme is implemented to ensure stable variable selection and to derive data-driven control thresholds for detecting out-of-control signals. The effectiveness of the framework is evaluated using five industrial case studies spanning low- to high-dimensional response settings and benchmarked against established MTS variants. Findings The empirical results indicate that the proposed framework achieves comparable or improved monitoring performance while substantially reducing dimensionality of responses. The results further demonstrate enhanced robustness and selection stability across diverse datasets relative to existing MTS-based approaches. Research limitations/implications The evaluation is based on five retrospective datasets drawn from the literature and real-world applications, which may limit the generalisability of the findings across broader industrial contexts. Future research should examine ultra-high-dimensional settings in which the number of in-control observations is smaller than the number of response variables, and further investigate the scalability and theoretical properties of the proposed framework. Originality/value This study introduces a fully integrated, distribution-free multi-objective optimization framework for variable selection and threshold determination within the Mahalanobis–Taguchi System for multivariate process control. By jointly optimizing competing objectives and incorporating selection stability via nested cross-validation, the proposed approach provides a practical, scalable solution for MPQC in complex, non-normal, and high-dimensional settings.