Machine Learning-Driven Best–Worst Method for Predictive Maintenance in Industry 4.0

The rapid proliferation of Industry 4.0 technologies has created an urgent need for intelligent and reliable predictive maintenance (PdM) systems. While multi-criteria decision-making (MCDM) frameworks like the Best–Worst Method (BWM) offer structured approaches for prioritizing maintenance tasks, their traditional reliance on subjective expert opinion limits their scalability and adaptability in dynamic industrial settings. This study addresses these limitations by introducing a robust, data-driven framework that integrates machine learning (ML) with BWM. This study presents a framework integrating ML models with BWM, an MCDM technique. While prior work has explored ML for fault detection/classification and hybrid MCDM + ML approaches, our innovation lies in automating BWM weight calculation via ML-derived feature importances, transforming tacit expert knowledge (traditionally subjective) into explicit, data-driven criteria weights aligned with Knowledge Management (KM) principles. The proposed methodology moves beyond a single-model proof-of-concept to present a comprehensive validation blueprint for industrial deployment. The framework’s efficacy is demonstrated using the standard Case Western Reserve University (CWRU) dataset, where rigorous cross-validation and statistical significance testing identified the optimal model, offering a compelling balance of high stability and efficiency for adaptive systems. Furthermore, simulations demonstrated the framework’s real-time viability, with low processing latency, and its resilience to concept drift through an adaptive retraining strategy. By integrating the empirically validated model’s feature importances into the BWM, this work establishes an objective, data-driven, and adaptive system for prioritizing maintenance, thereby advancing the transition toward autonomous and self-optimizing industrial ecosystems.