Application of multivariate adaptive regression splines technique for blast-induced air overpressure control in mines using large dataset

Rock mass in open cast mines is generally fragmented by blasting. The sudden release of explosive energy not only fragments rock but also generates undesirable nuisances such as ground vibrations, fly rock, dust, toxic fumes, back break, air overpressure (AOp) etc. AOp can vibrate civil engineering structures, shatter glass windows and doors, and adversely affect living quality. The present study strives to develop a reliable artificial intelligence model (AI) for predicting AOp in some Indian mines. For the objective, a large data set consisting of 699 data-points were collected from 33 open-cast mines to develop two white box AI techniques (multivariate adaptive regression splines and classification and regression tree), one black-box AI technique (support vector regression), multiple linear regressions, and USBM empirical predictor. The models were trained and tested using 70% (489) and 30% (210) datasets. Ten inputs, namely, hole diameter, hole depth, number of holes, burden, spacing, stemming length, charge per hole, total charge, maximum charge per delay, and distance, were used to develop these models. The performance of models was assessed using coefficient of determination (R2) and, root means squared error (RMSE). The results showed that the multivariate adaptive regression splines (MARS) model outperformed other models in predicting AOp with an R2 of 0.926, RMSE of 1.822. The proposed MARS model with a large dataset should better result in efficient prediction of AOp and can be easily implemented in different geo-environments for impacts minimization.