Intelligent prediction of blast-induced peak particle velocity (PPV) using an ensemble based multilayer machine learning approach for safe mining operations

Blasting is an important parameter in the assessment of structural safety and environmental impact in the mining and construction business. The prediction of peak particle velocity (PPV) is important to prevent damage to adjacent infrastructure. This study proposes a stacking ensemble based multilayer hybrid machine learning (ML) framework for enhancing the accuracy of blast-induced PPV prediction. The first layer of the model takes the outputs of several base learners such as Random Forest (RF), AdaBoost, XGBoost and CatBoost, combines them and passes them into a Gradient Boosting Regressor as a meta-learner in the second layer. The dataset consists of 993 samples including geometric parameters of the blast (blast-hole depth, spacing, burden, stemming and decking), explosive parameters (type, volume and powder factor (PF)) and operational parameters (average charge per hole (Avg. CPH), maximum charge per delay (MCPD) and seismograph distance). To avoid overfitting and underfitting and to improve the generalization capability of the models, a comprehensive hyperparameter optimization process was implemented. The comparison with single regressors, Bagging and Boosting models shows that the proposed stacking ensemble method achieves better predictive accuracy and reliability. The proposed model has been performing very well with R² (0.9855), RMSE (0.0178), MSE (0.00031), MAE (0.0079), SSE (0.0633), NRMSE (0.0246), MAD (0.0079) and VAF (98.52%). These results verify the effectiveness and practical utility of the proposed framework for reliable PPV prediction and provide valuable insights for optimizing blast design and vibration control in field operations.