The Evaluation of the Effects of Regulatory Variables on the Gravity of Coal Mine Disasters: A Machine Learning and SHAP Analysis Based Method

Abstract

Accurate assessment of accident severity and regulatory deficiencies is essential for improving safety management and intelligent decision-making in complex industrial systems. This study proposes an interpretable machine learning framework to evaluate the influence of regulatory variables on coal mine accident severity by integrating data-driven prediction with SHAP-based feature attribution. Based on grounded-theory analysis of 382 accident investigation reports, more than 800 textual statements were condensed into 62 initial concepts, 13 secondary indicators, and five categories of regulatory factors, including design schemes, management systems, technical documents, organizational measures, and process methods. These variables were encoded and incorporated into Logistic Regression, C4.5, CART, CHAID, and Random Forest models for comparative analysis. Experimental results demonstrate that the Random Forest model achieves the best predictive performance in terms of accuracy and AUC, while SHAP analysis provides quantitative interpretation of the contribution and interaction of regulatory variables. The findings indicate that organizational measures, technical documentation, process methods, and management systems are the dominant determinants of accident severity, enabling transparent risk assessment and targeted intervention strategies. The proposed framework establishes an effective methodology for interpretable predictive analytics, intelligent safety monitoring, and data-driven decision support in complex engineering environments, offering valuable references for distributed sensing systems, industrial information fusion, and intelligent monitoring architectures related to Electromagnetic Waves, Antennas and Propagation engineering applications.