A novel ensemble machine learning model to predict mine blasting-induced rock fragmentation

Abstract

In production blasting, the primary goal is to produce an appropriate fragmentation, whereas an improper fragmentation is one of the most common side effects induced by these events. This investigation aims at predicting rock fragmentation through a new ansemble technique, namely light gradient-boosting machine (LightGBM) with its hyper-parameters that were tuned using a powerful optimization algorithm, i.e., the Jellyfish Search Optimizer (JSO). The hybrid JSO-LightGBM is responsible for obtaining the highest possible performance from a combination of these two models where the used database is collected from the Sungun copper mine, Iran. Some blasting pattern parameters such as stemming and spacing were used as input variables while the mean fragment size (D50), which is a valid indicator for rock fragmentation studies, was considered an output variable. As a result, the coefficient of determination (R2) of 0.990 on the training set and R2 of 0.996 on the testing set confirmed that the newly developed JSO-LightGBM model has a powerful capability for predicting rock fragmentation, and it can be used as a new methodology in this field. Furthermore, the correlations between the input variables and target output by the Shapley Additive exPlanations technique showed that the powder factor has the most significant impact on fragmentation.