Machine learning models for prediction of blasting induced ground vibrations in basaltic rocks: a case study from Navi Mumbai Airport quarry, India

Abstract

Ground vibrations are a deleterious consequence of blasting. Peak particle velocity is used to assess the strength of the ground vibrations, and it usually estimated before the blasting operations. Traditional ground vibration predictors simplify the estimation by considering the maximum charge per delay and the distance from the blast site. Thus, the combined effect of the blast design parameters on the peak particle velocity is disregarded. Artificial intelligence algorithms can be used for the prediction of peak particle velocity considering the collective effects of the blast design parameters. The present research evaluates the different regression algorithms for prediction of the peak particle velocity of blast-induced ground vibrations. Feature used are the ratios of: spacing and burden, stemming and burden, blast hole depth and burden, burden and diameter, charge factor, maximum charge per delay, and the distance of the monitoring station and the target variable is peak particle velocity. A dataset consisting of 418 blasts carried out at the site in Western India. Thereafter, the supervised viz. Extra Tree, Extra Gradient Boost, Random Forest, Light Gradient Boosting Machine, Decision Tree, Support Vector have been adopted to predict the target variable. The results indicate that all the above methods have predicted the peak particle velocity with reasonable accuracy. It further shows that the Extra Tree regressor exhibits highest R2 score of 0.88, for the test set, with a RMSE of.434. The study concludes that tree-based ensemble techniques can be used for reliable prediction when the data set is limited. © Springer Nature Switzerland AG 2025.