TITLE:
Performance Evaluation of Soft Computing Techniques in Blast-Induced Predictions: The Case of Backbreak
AUTHORS:
Festus Kunkyin-Saadaari, Charles Asiedu, Ephraim Atta-Duncan, Victor Kwaku Agadzie, Mary Adu-Gyamfi
KEYWORDS:
Backbreak Prediction, Machine Learning, Open-Pit Mining, Gradient Boosting, Neural Networks, Blast Design Parameters
JOURNAL NAME:
International Journal of Geosciences,
Vol.16 No.10,
October
29,
2025
ABSTRACT: The accurate prediction of backbreak, a crucial parameter in mining operations, has a significant influence on safety and operational efficiency. The occurrence of this phenomenon is detrimental to the safety, capital and human resources of a mine. This framework applies machine learning algorithms to predict backbreak. An enhanced precision will be explored specifically employing gradient boosting decision trees (GBDT), light gradient boosting machines (LightGBM), backpropagation neural network (BPNN), Graph Neural Networks (GNNs) and Kolmogorov-Arnold Network (KAN) algorithm. The study utilises a comprehensive dataset from the Goldfields Ghana Limited, Damang Mine comprising geomechanical, drilling, and blasting parameters (burden, spacing, stemming height, geometric stiffness, and powder factor) as well as backbreak data. The potential of each algorithm to learn the intricate relationships between the input features and backbreak values is investigated. To quantitatively assess the predictive performance of the models, the evaluation metrics, coefficient of determination (R2), mean absolute error (MAE), and mean square error (MSE) are employed. The results revealed that GBDT and BPNN algorithms exhibited robust predictive capabilities, capturing the complex non-linear patterns in the dataset, achieving higher R2 values (97% and 92% respectively) and lower MAE scores (0.2603 and 0.456, respectively) and MSE scores (0.1456 and 0.3798, respectively). The LightGBM and KAN models also showed their predictive potential and captured the complex non-linear patterns in the dataset but not as efficiently as GBDT and BPNN. GNN showed the least performance in backbreak prediction. The findings highlighted the potential of the GBDT model to enhance backbreak prediction accuracy, thereby aiding in safer and more efficient excavation practices.