Abstract：The rock burst is affected by many complex factors，so the forecast of the degree of the rock burst is a nonlinear，high dimensional，multiclass pattern recognition with small samples. Based on statistical learning theory and complying with the minimization of structure risk，a new machine learning tool—support vector machine，which can solve the problems for multidimensional functions and has good extrapolating ability at small samples occasions，and that fetches up the ANN¢s insufficiencies，is employed. In order to improve the training velocity and prediction accuracy，this paper presents a new method for forecasting rock burst based on least square support vector machine，and constructs the prediction model. The rock burst¢s influence factors are mining depth，having or no pillar coal，rock character of top plate，intricacy degree of architectonic state，coal seam pitch，thickness of coal seam，mining system，workface by blasting or vertical exploitation. The complicated nonlinear relationship between the degree of rock burst and its affected factors is presented. The application to the practical engineering shows that the method is feasible and precise.
