Rock burst grade prediction based on KPCA-BAS-SVM model

With the development of mines, tunnels and other underground projects to the deep, the accurate prediction and early warning of rock burst disasters can effectively reduce the damage to construction personnel and equipment. Based on the causes, characteristics, and influencing factors of rock bursts, this study selects maximum tangential stress （MTS）, uniaxial compressive strength （UCS）, uniaxial tensile strength （UTS）, stress coefficient （SCF）, brittleness index （BI）, and elastic energy index （WET） as key indicators for rock burst prediction. A rock burst database is established based on 184 case studies collected from both domestic and international sources. To simplify model input parameters while preserving essential features of the rock burst data, a linear weighted fusion strategy is adopted, combining polynomial kernel function with Gaussian radial basis function to construct a hybrid kernel function, and grid search is used to determine their optimal combination coefficients. The antlion optimizer algorithm （BAS） is employed to optimize the penalty parameter c and kernel parameter g in support vector machines （SVM）, eliminating subjective human interference that may affect the reliability of rock burst predictions. The optimized model is trained and tested using three kernel principal components derived from dimensionality reduction. Model performance is evaluated using confusion matrices and multiple evaluation metrics, and the results are compared with those of conventional SVM and BAS-SVM models. Results show that the KPCA-BAS-SVM model achieves an accuracy rate of 89.3%, representing improvements of 17.4% and 13.6% over the SVM and BAS-SVM models, respectively. Furthermore, the model demonstrates average precision of 91.0%, average recall of 91.7%, and an F1 score of 91.3%, all significantly outperforming the comparison models. When applied to Kuocangshan Tunnel and Maluping Mine, the predicted rock burst levels closely match actual conditions, verifying the model’s feasibility and applicability. This provides a new approach for rock burst early warning.