An Information Entropy–based Risk (IER) Index of Mining Safety Using Clustering and Statistical Methods

Abstract

In recent decades, the mining industry in the United States has made significant strides in reducing accidents and injuries. While these improvements are commendable, interpreting these statistics can be challenging due to concurrent declines in workforce size, employee hours, productivity, and operating systems. The Mine Safety and Health Administration (MSHA) of the United States has instituted tools like the Pattern of Violation (POV) and Significant & Substantial (S&S) calculator to monitor safety in mines. However, both have their respective limitations. Various risk indices have been proposed to address these limitations, leveraging multiple matrices from MSHA databases. Yet, the primary challenge lies in effectively integrating these diverse matrices into a cohesive risk index. This research endeavors to develop an information entropy–based risk (IER) index through the optimization of weights assigned to these sometimes-conflicting matrices. The seven-dimensional risk indicators considered for IER index computation encompass (a) citations, (b) orders, (c) significant & substantial citations, (d) penalties, (e) incidents with no lost time, (f) lost time injuries, and (g) proposed penalty for violation. The efficacy of the proposed IER index was assessed using data from MSHA’s underground mines spanning from 2011 to 2020. Validation of the IER index was conducted through application of the BIRCH clustering algorithm in tandem with rigorous statistical analysis. The clustering performance was evaluated using the multivariate analysis of variance (MANOVA) test, followed by post hoc analysis. Box plots and univariate analysis of variance (ANOVA) tests were then employed to substantiate the statistical significance of mean differences in IER index values across clusters. The MANOVA test and subsequent post hoc results underscore the successful clustering of the seven-dimensional risk indices across all time periods using the BIRCH algorithm. The ANOVA test unequivocally demonstrates that the mean risk index values of at least one cluster are statistically distinct from the others at a 95% confidence level for all periods. Post hoc analysis further confirms the statistical significance of differences in mean risk indices between clusters. These findings were further supported by the results obtained from the box plots. Finally, the proposed approach was applied to an underground coal mine to illustrate its practical effectiveness. This study demonstrates that the proposed approach can empower mining companies to comprehensively assess their safety performance and implement necessary measures for improvement.