A three-step rockburst prediction model based on data preprocessing combined with clustering and classification algorithms

Currently, the rockburst failure mode discrepancy is not taken into account in rockburst prediction, and some defects such as missing values, outlier samples and class imbalance are still present in the rockburst database. To solve the above problems, a three-step rockburst prediction model based on data preprocessing combined with clustering and classification algorithms (PCC) is proposed. The first step is data preprocessing for dealing with missing values and outlier samples. In the second step, the K-means algorithm is introduced to cluster the preprocessed data into two clusters for distinguishing the rockburst failure mode discrepancy. In the third step, Random Forest (RF), Gradient Boosting Decision Tree (GBDT) and Extreme Gradient Boosting (XGB) are employed to construct classification models, respectively, after class balancing. Finally, a comparative analysis between the PCC model and the rockburst prediction model based on data preprocessing and classification algorithms (PC) is conducted, and the analysis is based on 310 samples. Furthermore, the self-proposed metric and traditional metrics are employed to evaluate the model performance. The prediction results demonstrate that the performance of the PCC model is superior to the PC model, and the rockburst failure mode discrepancy should be considered in rockburst prediction.