Suitability assessment of different vector machine regression techniques for blast-induced ground vibration prediction in Ghana

Abstract

There is a collective demand for the mining industry to accurately predetermine or predict blast-induced ground vibration to support effective blast control management. It is, therefore, desirable to explore and develop accurate forecasting tools that can meet environmental and safety standards. In this study, the prediction capabilities of two proposed models, namely least squares support vector machine (LSSVM) and relevance vector machine (RVM) were explored and compared with support vector machine (SVM) which has found a wide application in blast-induced ground vibration prediction. The prediction results of these techniques were ranked to identify the best using mean square error (MSE), root-mean square error (RMSE), correlation coefficient (R) and model efficiency of Loague and Green (ELG). The ranking results revealed that LSSVM was superior to the SVM and RVM. The results produced by LSSVM model were further compared to three benchmark ANN methods (backpropagation neural network (BPNN), radial basis function neural network (RBFNN) and generalised regression neural network (GRNN)) and five empirical predictor models (United State Bureau of Mines (USBM), Ambraseys-Hendron, Langefor-Kihlstrom, Indian standard and Central Mining Research Institute (CMRI)). The comparative analysis revealed that the LSSVM was the best prediction approach because it achieved the lowest MSE and RMSE results and recorded the highest R and ELG values of 0.0215, 0.1467, 0.8542 and 0.7273, respectively. The application of Bayesian Information Criterion (BIC) for model selection confirmed LSSVM as the best among all the methods applied because it produced the least BIC value of − 285.2312.