Computational Intelligence for Blast-Induced Backbreak Prediction in High-Stress Mining Environments

Ensuring the stability of mine walls and floors is essential for maintaining pit safety and maximizing productivity. One of the key challenges in surface mining is predicting blast-induced backbreak, a phenomenon directly influenced by blast design parameters and the efficient use of explosive energy. This study presents advanced hybrid models based on artificial intelligence (AI) to forecast blast-induced backbreak in surface mines based on a database from two major mining operations located in the southwest region of Zanjan Province, Iran. A suite of machine learning algorithms was explored, including extreme gradient boosting (XGBoost), CatBoost, AdaBoost, classification and regression trees (CART), random forest (RF), decision trees (DT), extra trees (ET), gradient boosting machines (GBM), and light gradient boosting machines (LightGBM). These models were further optimized using the Jellyfish Search Optimizer (JSO), with performance evaluated through multiple statistical indicators. The study analyzed data from 759 blast events, incorporating input variables such as hole length (L), burden (B), spacing (S), hole angle (H), stemming length (St), and charge per delay (CD). Among all tested models, the JSO-optimized XGBoost (JSO-XGBoost) demonstrated superior performance, achieving a Mean Absolute Error (MAE) of 0.104, a coefficient of determination (R 2 ) of 0.984, a Root Mean Square Error (RMSE) of 0.122, a Variance Accounted For (VAF) of 98.33%, and a bias of 0.045. Among the tested models, the JSO-XGBoost hybrid model exhibited the highest predictive performance. This superiority is attributed to its enhanced capability to capture complex nonlinear relationships among blasting parameters and its robustness to noise and variability in the dataset, making it particularly effective for accurate backbreak prediction. These results highlight the effectiveness of AI-driven hybrid modeling, particularly the JSO-XGBoost approach, in accurately predicting blast-induced backbreak and supporting improved stability assessments of mine walls and floors. • Proposes a novel Jellyfish Search Optimizer (JSO)-enhanced hybrid AI approach to predict blast-induced backbreak, advancing computational rock mechanics. • Evaluates nine machine learning models using 759 field blast cases, with JSO-XGBoost demonstrating superior accuracy (R 2 = 0.984, VAF = 98.33%). • Provides actionable insights for optimizing blast designs to mitigate rock damage and enhance pit stability in high-stress mining environments. • Leverages real-world data from active Iranian mines, ensuring robust applicability to extreme geotechnical conditions.