A Hybrid Model for Back-Break Prediction using XGBoost Machine learning and Metaheuristic Algorithms in Chadormalu Iron Mine

Back-break is one of the adverse effects of blasting, which results in unstable mine walls, high duration, falling machinery, and inappropriate fragmentation. Thus, the economic benefits of the mine are reduced, and safety is severely affected. Back-break can be influenced by various parameters such as rock mass properties, blast geometry, and explosive properties. Therefore, during the blasting process, back-break must be accurately predicted, and other production activities must be done to prevent and reduce its adverse effects. In this regard, a hybrid model of extreme gradient boosting (XGB) is proposed for predicting back-break using gray wolf optimization (GWO) and particle swarm optimization (PSO). Additionally, validation of the hybrid model is conducted using XGBoost, gene expression programming (GEP), random forest (RF), linear multiple regression (LMR), and non-linear multiple regression (NLMR) methods. For this purpose, the data obtained from 90 blasting operations in the Chadormalu iron ore mine are collected by considering the parameters of the blast pattern design. According to the results obtained, the performance and accuracy level of hybrid models including GWO-XGB (R2 = 99, RMSE = 0.01, MAE = 0.001, VAF = 0.99, a-20 = 0.98), and PSO-XGB (99, 0.01, 0.001, 0.99, 0.98) are better than the XGBoost (97, 0.185, 0.132, 0.98, 95), GEP (96, 0.233, 0.186, 0.967, 0.935), RF (97, 0.210, 0.156, 0.97, 0.94), LMR (96, 0.235, 0.181, 0.964, 0.92), and NLMR (96, 0.229, 0.177, 0.968, 0.93) models. Notably, the GWO-XGB hybrid model has superior overall performance as compared to the PSO-XGB model. Based on the sensitivity analysis results, hole depth and stemming are the essential effective parameters for back-break.