Prediction of Blasting Efficiency in Mining: A Comprehensive Evaluation with Boosting-Based Machine Learning Algorithms

OpenAlex Turkish

This study focuses on predicting blasting efficiency in underground mining operations using real-world operational data. The primary objective is to accurately estimate blasting efficiency through regression analysis, leveraging the predictive power of state-of-the-art boosting-based machine learning (ML) algorithms. Given the relatively small and multi-variate nature of the dataset, algorithms with low overfitting risk, flexibility, and high speed were preferred. We comparatively evaluate five different cutting-edge boosting algorithms: Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Natural Gradient Boosting (NGBoost), and Histogram-based Gradient Boosting Machine (HistGBM). Additionally, to further enhance prediction performance, three distinct ensemble learning strategies—Simple Averaging, Weighted Averaging, and Weighted Average with the Best 3 Models—were implemented and tested. The methodology involved comprehensive data preprocessing, including the removal of irrelevant variables like Date, numerical transformation of categorical features using Target Encoding, and normalization of numerical inputs with StandardScaler. The dataset, comprising 652 observations from 2019-2020 underground blasting operations, was split into 80% training and 20% testing sets, and 10-Fold Cross Validation was employed for model training. Hyperparameter optimization for each boosting algorithm was performed using manual tuning, GridSearchCV, and RandomizedSearchCV. Model performance was assessed using R2, Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE) metrics for both cross-validation and test sets. Our findings indicate that ensemble methods significantly improve prediction accuracy. The comparative analysis allowed for the identification of the most suitable machine learning approach for blasting efficiency prediction. This research contributes significantly to the scientific literature by providing a robust framework for enhancing blasting efficiency and optimizing operational processes through data-driven decision support systems in mining environments. The results demonstrate the effectiveness of combining multiple boosting algorithms and ensemble techniques for accurate and reliable blasting efficiency prediction.