A hybrid machine learning framework for multidimensional risk analysis with integrated uncertainty quantification and explainability for optimized inspection planning

Purpose Balancing cost-effectiveness and availability against safety and environmental stewardship remains a persistent challenge in asset management. Conventional Risk-Based Inspection (RBI) often depends on imprecise qualitative assessments, while modern data-driven methods frequently lack transparency. This study addresses this gap by introducing the multidimensional risk analysis (MDRA) framework, which offers a comprehensive, transparent, and actionable risk profile for safety-critical assets. Design/methodology/approach A hybrid intelligence system utilizing a weighted, stacked ensemble classifier was developed to predict the probability of failure (PoF). Bootstrap resampling was employed to generate 95% confidence intervals for uncertainty quantification. To address the limitations of the “black box” nature of standard AI, SHapley Additive exPlanations (SHAP) were integrated to ensure full model interpretability. Finally, a hierarchical, rule-based engine translated these probabilistic outputs into actionable inspection plans for industrial assets. Findings The application of the MDRA framework successfully demonstrated the capacity to generate reliable and explainable inspection priorities. By integrating deterministic rules with probabilistic machine learning insights, the system produced robust, data-driven inspection plans that accounted for uncertainty. The framework proved capable of identifying the equilibrium between six critical operational values: safety, integrity, reliability, availability, environmental responsibility, and cost-effectiveness, translating complex risk data into clear, actionable maintenance decisions. Originality/value This paper presents a new framework that overcomes key limitations of traditional risk-based methods and opaque advanced analytics. Its main contribution is the integration of explainable AI (XAI) and uncertainty quantification into safety-critical asset management. By combining model interpretability with probabilistic risk estimates, the MDRA framework improves the transparency, robustness, and reliability of data-driven risk assessments. This dual focus on explanation and uncertainty allows for more confident and context-aware prioritization of inspection and maintenance activities, supporting operational efficiency while upholding safety and environmental standards.