Flyrock distance prediction using a hybrid LightGBM ensemble learning and two nature-based metaheuristic algorithms

Traditional mining in open pit mines often uses explosives, leading to environmental hazards, with flyrock being a critical issue. In detail, excess flying rock beyond the designated explosion area was identified as the primary cause of fatal and non-fatal blasting hazards in open pit mining. Therefore, the accurate and reliable prediction of flyrock becomes crucial for effectively managing and mitigating associated problems. This study used the Light Gradient Boosting Machine (LightGBM) model to predict flyrock in a lead-zinc mine, with promising results. To improve its accuracy, multi-verse optimizer (MVO) and ant lion optimizer (ALO) metaheuristic algorithms were introduced. Results showed MVO-LightGBM outperformed conventional LightGBM. Additionally, decision tree (DT), support vector machine (SVM), and classification and regression tree (CART) models were trained and compared with MVO-LightGBM. The MVO-LightGBM model excelled over DT, SVM, and CART. This study highlights MVO-LightGBM's effectiveness and potential for broader applications. Furthermore, a multiple parametric sensitivity analysis (MPSA) algorithm was employed to specify the sensitivity of parameters. MPSA results indicated that the highest and lowest sensitivities are relevant to blasted rock per hole and spacing with the γ = 1752.12 and γ = 49.52, respectively.