Fostering sustainable mining practices in rock blasting: Assessment of blast toe volume prediction using comparative analysis of hybrid ensemble machine learning techniques

Abstract

Blast toe volume, pivotal in explosive engineering, underpins explosive energy efficient utilization, blast safety and mine production sustainability. While current research explores the use of artificial intelligence (AI) model to maximize toe volume prediction, gaps persist in understanding the application of ensemble learning algorithm techniques like hybrid and voting techniques in addressing toe volume problem. Bridging these gaps promises enhanced safety and optimization in blasting operations. This study Performs AI model hybrid and voting to enhance toe volume prediction model robustness by leveraging diverse algorithms, mitigating biases, and optimizing accuracy. The study combines separate models, looks for ways that hybrid approaches can work together, and improves accuracy through group voting in order to give more complete information and more accurate predictions for estimating blast toe volume in different approaches. To develop the models, 457 blasting data was collected at the Anguran lead and zinc mine in Iran. The accuracy of the developed models was assessed using nine indices to compare their prediction performance. To understand the input relationship, multicollinearity, spearman, pearson, and kendall correlation analyses show that there is a strong link between the size of the toe and the explosive charge per delay. Findings from the model analysis showed that the Light Gradient Boosting Machine (LightGBM) was the most accurate of the eight traditional models, with R2 values of 0.9004 for the training dataset and 0.8625 for the testing dataset. The Hybrid 6 model, which combines LightGBM and Classification and Regression Trees (CART) algorithms, achieved the highest R2 scores of 0.9473 in the training phase and 0.9467 in the testing phase. The Voting eight models, consisting of LightGBM, Gradient Boosting Machine (GBM), Decision tree, Ensemble tree, Random Forest, CatBoost, CART, AdaBoost, and XGBoost, had the greatest R2 scores of 0.9876 and 0.97265 in both the training and testing stages. Using novel modelling tools to forecast blast toe volume in this study allows for resource extraction optimization, decreases environmental disturbance through mine toe smoothening, and improves safety, supporting sustainable mining practices and long-term sustainability.