Rock burst intensity grading prediction based on the combination of PCA，CBLOF and SVMSMOTE algorithms

To reduce the correlation among feature parameters in rock burst sample data，eliminate outliers and balance the number of various rock burst levels in the dataset，a method combining principal component analysis (PCA)，cluster-based local outlier factor(CBLOF) and support vector machine synthetic minority over-sampling technique(SVMSMOTE) is proposed to enhance the quality of the rock burst database. Initially，a total of 343 rock burst cases are collected from both domestic and international sources to build the original dataset. PCA is employed for dimensionality reduction，CBLOF is used to identify and eliminate outliers within each rock burst level，and SVMSMOTE synthesizes new minority samples near the boundaries of each rock burst level. The processed and original rock burst databases are subsequently used to train six different machine learning models to validate the effectiveness of the PCA，CBLOF，and SVMSMOTE combination. The results show that the accuracy of models trained on the processed database significantly improved：AdaBoost by 29%，CatBoost by 28.5%，LightGBM by 34%，Gradient Boosting by 28%，ExtraTrees by 26.5% and Random Forest by 24%，compared to models trained on the original database. Therefore，processing the rock burst database using the combined PCA，CBLOF and SVMSMOTE algorithms effectively enhances the quality of the database and improves the performance of machine learning prediction models.