Developing intelligent classification models for rock burst prediction after recognizing significant predictor variables, Section 2: Designing classifiers

Rock burst phenomenon prediction is one of the important issues in the mining, geo-mechanics and strata control fields. Predictive classification consists of constructing models designed with the goal of exact and reliable forecasting the values of nominal target variables using information on a set of predictor variables. In this research, after regarding the particular considerations of the classification problem and identifying the key and effective predictor variables by summing up the results of the data preprocessing procedure on a rock burst database, some of the most effective classification methods currently known with totally different attitudes for separating the decision space, including five widespread families of the naive Bayesian, decision tree, nearest neighbor, support vector machine and artificial neural network techniques, were employed to predict the occurrence and intensity of the phenomenon in the form of distinct problems. The models with the highest classification accuracies were recognized based on the considered error estimators adopting different estimation strategies. These paramount classification models had satisfactory outcomes and also significantly higher performances than the simultaneous application of the empirical criteria presented based on the in-review predictor variables.