Abstract: To predict rock burst risk classification accurately, an optimized Bagging-LSSVM prediction model was built. On the basis of designing and optimizing the algorithm flow of Bagging-LSSVM, a set of classical classification datasets was introduced to the experiment. A minimum number was obtained experimentally for basic classification model's number in meetting optimal classification. Main factors affecting rock burst were identified to futher determine the evaluation indexes. Then 35 groups of measured data provided by the Chongqing Yanshitai mine were used, as samples to test. Correlations among original sample indexes were eliminated by using KPCA. Then the rock burst prediction accuracy comparison was made among the optimized Bagging-LSSVM model, optimized Bagging-SVM model and LSSVM model. It is turned out that forecasting accuracy by the optimized Bagging-LSSVM model is greater than those by others.

Key words: rock burst, risk, least squares support vector machine(LSSVM), optimized Bagging-LSSVM, kernel principal component analysis(KPCA)