Predicting Blast-Induced Ground Vibrations in Some Indian Tunnels: a Comparison of Decision Tree, Artificial Neural Network and Multivariate Regression Methods

The present study compares three different techniques (decision tree, artificial neural network and multivariate regression analysis) for predicting blast-induced ground vibrations in some Indian tunnelling projects. The performance of these models was also compared to site-specific conventional predictor equations. A database consisting of 137 vibration records was randomly divided into training and testing sets for model generation. Eight input parameters (total charge, tunnel cross-section, maximum charge per delay, number of holes, hole diameter, distance from blasting face, hole depth and charge per hole) were selected for model development using bivariate correlation analysis. Results indicated that the decision tree is best suited for predicting vibrations. The decision tree further suggested that the intensity of near-field ground vibrations is mainly affected by total charge fired in a round, whereas the intensity of far-field vibrations is governed by maximum charge per delay and charge per hole. Conventional ground vibration predictors and machine learning techniques such as neural networks do not depict the relationship between input and output parameters. However, the present study substantiates that the decision tree can be a good tool for precise prediction of ground vibrations. Further, the decision tree can classify and relate different blast design parameters for refining blast designs to control ground vibrations on sites.