Forecast of mining fatalities using machine learning algorithms

Mining industry plays a crucial role in economic development but is often accompanied by notable occupational hazards, including fatalities. This research investigates the application of machine learning techniques to predict mining fatalities in the United States using historical data from 1983 to 2023, sourced from the National Institute for Occupational Safety and Health (NIOSH) and FRED Economic Data. Predictive models including Random Forest (RF), Gradient Boosting (GB), Decision Tree (DT), Multiple Linear Regression (MLR), and Linear Regression (LR), were employed to assess relationships between operational parameters and annual fatality counts. The dataset was thoroughly preprocessed, addressing missing values and deriving a fatality rate metric to improve predictive accuracy. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the Coefficient of Determination (R²). Among the models, GB achieved the highest predictive accuracy, with an MAE of 7.85, RMSE of 10.04, and R² of 0.903, followed closely by RF and LR approaches. These findings highlight the effectiveness of machine learning, especially ensemble methods, in modeling complex patterns and offering valuable insights for improving safety outcomes in the mining sector. To support interpretability, SHapley Additive exPlanations (SHAP) analysis was employed to explain feature contributions, providing insights into how key variables influenced model predictions.