Abstract: In order to further improve the prediction accuracy of the rock burst risk, the ensemble learning (EL) method was used to analyze the main factors and indicators of the occurrence of rock burst. Seven kinds of classification prediction models in the EL method were respectively used to predict the rock burst risk. The experimental results show that all the seven models have certain reliability. Taking the accuracy and Haiming loss of the models as evaluation indexes, it is concluded that the XGBoost algorithm has high prediction performance and can predict the rock burst risk relatively effectively. Finally, the SHAP value is used to further explain the XGBoost model. The elastic energy index has the greatest influence on the rock burst risk.