Hybridization of Parametric and Non-parametric Techniques to Predict Air Over-pressure Induced by Quarry Blasting

This study employed a hybridization approach that combines parametric and non-parametric models to predict air over-pressure (AOp) associated with quarry blasting. A simple linear regression model, which is a kind of parametric model, was used to select the most relevant inputs for predicting AOp. Four parametric models, including Chi-square automatic interaction detector (CHAID), artificial neural network (ANN), k-nearest neighbors (KNN), and support vector machine (SVM) were developed using the outputs of a linear model to predict AOp. The models developed were evaluated using five performance indicators, a simple ranking system, and a gains chart. According to the evaluations, ANN and CHAID (both with cumulative ranking = 36) outperformed SVM (cumulative ranking = 15) and KNN (cumulative ranking = 24) to predict AOp. While CHAID (training ranking = 20) performed better than other models in the training phase, ANN (testing ranking = 20) performed better than the other models in the testing phase. In addition, while ANN and CHAID models identified distance as the least important factor for predicting AOp, there was no agreement on the most important factor. Moreover, a comparison between the present study and other studies that used the same dataset showed that, compared to the hybridization of non-parametric models, the hybridization of parametric and non-parametric models potentially results in better accuracy.

Automated summary

This study employed a hybridization approach to predict air over-pressure associated with quarry blasting, with artificial neural network and Chi-square automatic interaction detector performing the best. Distance was identified as the least important factor for predicting AOp, while the most important factor remains inconclusive.