Modern Approaches to Air Overpressure Prediction: Evaluating Ensemble Machine Learning and Classical Statistical Models

Mining provides essential raw materials for various sectors but carries significant risks due to hazardous processes. Taking valuable minerals or other geological materials out of the earth is known as mining. Resources like coal and metals, placer, underground, and surface mining are essential, but they also have negative environmental effects, such as air pollution from blasting and water pollution. Air noise, frequently caused by industrial operations like mining and construction, can harm wildlife and human health. Transportation, equipment, and blasting activities are examples of sources. To reduce the negative effects of high noise levels on the environment, stress, and hearing loss, noise management and predictive models are crucial by establishing correlations between variables such as charge weight, distance, and geological conditions. Statistical predictor equations calculate blast-induced Air Overpressure (AOp). In India, DGMS regulations ensure mining and blasting operations minimise environmental impacts and keep AOp levels safe for nearby communities. In this study, SVR, RF, GB, BPNN, and an ensemble hybrid XGBoost–RF model were developed to predict blast-induced AOp and compared with traditional statistical prediction equations. The performance of these models was evaluated using four metrics: RMSE, MSE, MAE, and R². The results showed high accuracy for machine learning models, with R² values up to 0.9991 for the ensemble hybrid model, compared to much lower R² values for classical statistical approaches. These findings demonstrate the effectiveness of modern machine learning methods in predicting blast-induced Air Overpressure and highlight their superiority over traditional statistical models.