Ensemble machine learning models for prediction of flyrock due to quarry blasting

In the mining industry, the most common approach to rock fragmentation is blasting. Blasting operations generate flyrock, which is a critical and tough task, and its assessment is critical in decreasing related hazards. In this study, ensemble learning approaches such as simple averaging ensemble, weighted averaging ensemble, integrated stacking model, separate stacking model, and Bayesian-eXtreme Gradient Boosting are used to establish a predictive model for the flyrock generated by blasting. This effort resulted in a separate stacking model with a bagging met-learner, which overall outperforms other models. The mean square error, the coefficient of determination, and the coefficient of variation for this model are 0.0059, 0.974, and 0.22, respectively. The SHapley Additive exPlanations (SHAP) methodology is employed to reveal the relative relevance of the parameters affecting the model's flyrock estimation. Based on the SHAP method, the hole diameter is determined as the main factor in controlling flyrock distance, which is followed by the powder factor, hole depth, and burden-to-spacing ratio. The framework and modeling process of this research would be useful for mining engineers/designers to minimize undesired environmental issues of blasting.