Abstract: Rockburst disasters occur frequently in mine roadways,traffic tunnels and other projects,and rockburst prediction has become particularly important. In order to improve the prediction accuracy of rockburst grade and the generalization of prediction model,an SSA-RF model using sparrow search algorithm (SSA) to optimize the random forest algorithm (RF) model was proposed. Considering the rationality of the rockburst grade prediction indicators,the four sets of schemes were selected based on the characteristics of rockburst genesis and the grey correlation analysis results to determine the best combination of prediction indicators,and the importance analysis of model was used to verify the rationality of the best combination of prediction indicators. The first scheme retained all the prediction indicators as a comparison item,the second scheme sieved out the two prediction indicators of low grey correlation with rockburst grade,the third scheme adopted composite indicators,and the fourth scheme adopted independent indicators. A total of 151 rockburst sample data were collected as model datasets,the prediction effect of SSA-RF model was compared with seven other prediction methods,and the sensitivity analysis of model for different sample sizes was analyzed. The results show that the second scheme is the best one,and the best combination of prediction indicators is tangential stress in surrounding rock, stress coefficient, elastic energy index and uniaxial compressive strength. Compared with the other seven prediction methods,the SSA-RF model has the highest prediction accuracy and the average accuracy reaches 88. 09% in the four sets of schemes,and even reaches 95. 23% in the best scheme. The prediction indicators of SSA-RF model are more reasonable in importance ranking than that of RF model,and effectively verify the rationality of the best combination of prediction indicators. The use of SSA algorithm to optimize the RF model can improve the prediction accuracy and generalization of the RF model for different rockburst sample volumes,and effectively compensate for the shortcomings of the randomness of the RF model. In summary,it can be seen that the SSA-RF model is feasible and effective for predicting the rockburst grade of actual engineering.

Key words: rockburst grade prediction,grey correlation analysis,SSA-RF model,importance analysis of model,sensitivity analysis of model