A Comprehensive Assessment of Mining Accident Severity Using Machine Learning Methods
A Comprehensive Assessment of Mining Accident Severity Using Machine Learning Methods

Summary/Abstract: The mining sector is recognized as one of the riskiest in the world. However, there is a lack of a comprehensive understanding of the factors influencing accident severity. In this study, eight machine learning (ML) methods were systematically evaluated, among which Gradient Boosting and Random Forest emerged as the top-performing models. These ensemble models demonstrated consistently high performance across multiple evaluation metrics (Accuracy, Precision, Recall, F1-score, and ROC AUC), highlighting their robustness and reliability in distinguishing between fatal and non-fatal accident outcomes. Across both impurity-based (Gini importance) and robust model-agnostic (SHAP) frameworks, Risk Taking, Age, and Experience emerge as the most influential predictors. Social engagement metrics vary in importance, collectively suggesting that social support systems may play a meaningful role in moderating the severity of accidents. These findings substantiate the utility of machine learning not only for accurate outcome classification but also for elucidating the multifaceted drivers of accident severity, thereby informing targeted intervention and prevention strategies.

• Details
• Contents