AI-Driven Predictive Maintenance: Bayesian Reliability Assessment and XGBoost for Smart Industrial Systems

This study proposes an interpretable predictive maintenance architecture that integrates XGBoost for remaining useful life prediction, Bayesian calibration for uncertainty assessment, and SHAP for transparent model explanation. Experiments conducted on benchmark degradation datasets demonstrate that the proposed approach achieves superior predictive performance while reducing computational burden when compared with commonly used baseline models under identical preprocessing conditions. The Bayesian component produces prediction intervals with empirical coverage close to the nominal confidence level, thereby enhancing the reliability of maintenance decisions. In addition, SHAP analysis reveals physically meaningful degradation drivers, including temperature, pressure, and vibration, which supports model auditability and facilitates root cause analysis. The primary contribution of this study lies in the unification of gradient boosted prediction, Bayesian reliability calibration, and explainable artificial intelligence techniques into a single reproducible pipeline that effectively translates predictive outputs into actionable maintenance scheduling insights.