Environment
Clement Kweku Arthur; Yao Yevenyo Ziggah; Victor Amoako Temeng
Abstract
Blast-induced noise is one of the most persistent environmental challenges in surface mining, posing significant health risks to workers and nearby communities. Accurate prediction of noise levels prior to blasting is essential for mitigating its adverse impacts. This study proposes an explainable ensemble ...
Read More
Blast-induced noise is one of the most persistent environmental challenges in surface mining, posing significant health risks to workers and nearby communities. Accurate prediction of noise levels prior to blasting is essential for mitigating its adverse impacts. This study proposes an explainable ensemble machine learning framework for predicting blast-induced noise using data from an open-pit gold mine in Ghana. Four ensemble models namely: Extreme Gradient Boosting (XGBoost), Gradient Boosting, Adaptive Boosting (AdaBoost), and Categorical Boosting (CatBoost), were developed and evaluated using a comprehensive dataset of 324 blasting events. Performances of the developed models were assessed using coefficient of determination (R²), root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and coefficient of the variation of the root mean squared error (CVRMSE), with XGBoost emerging as the best-performing model (R² = 1.0000, RMSE = 0.0005, MAE = 0.0004, MAPE = 0.0010, CVRMSE = 0.0013). To address the black-box nature of ensemble method, Shapley Additive exPlanations (SHAP) was employed, offering both global and local interpretability. SHAP analysis identified the distance from the blast site to the monitoring point as the most influential factor. This integrative approach not only enhances predictive accuracy but also improves model transparency, supporting sustainable mining practices aligned with United Nations Sustainable Development Goals (SDGs) 3 and 15.
