A comparative study of empirical and ensemble machine learning algorithms in predicting air over-pressure in open-pit coal mine

Abstract

This study aims to take into account the feasibility of three ensemble machine learning algorithms for predicting blast-induced air over-pressure (AOp) in open-pit mine, including gradient boosting machine (GBM), random forest (RF), and Cubist. An empirical technique was also applied to predict AOp and compared with those of the ensemble models. To employ this study, 146 events of blast were investigated with 80% of the total database (approximately 118 blasting events) being used for developing the models, whereas the rest (20% ~ 28 blasts) were used to validate the models’ accuracy. RMSE, MAE, and R2 were used as performance indices for evaluating the reliability of the models. The findings revealed that the ensemble models yielded more precise accuracy than those of the empirical model. Of the ensemble models, the Cubist model provided better performance than those of RF and GBM models with RMSE, MAE, and R2 of 2.483, 0.976, and 0.956, respectively, whereas the RF and GBM models provided poorer accuracy with an RMSE of 2.579, 2.721; R2 of 0.953, 0.950, and MAE of 1.103, 1.498, respectively. In contrast, the empirical model was interpreted as the poorest model with an RMSE of 4.448, R2 of 0.872, and MAE of 3.719. In addition, other findings indicated that explosive charge capacity, spacing, stemming, monitoring distance, and air humidity were the most important inputs for the AOp predictive models using artificial intelligence.