Abstract: In order to solve the problem that when the samples are few,the reliability of the prediction results of rockburst intensity classification obtained by machine learning algorithms is insufficient, a XGBoost algorithm based on multiple cross validation is adopted and the applicability is discussed. Firstly, five factors including the rock uniaxial compressive strength σc, the uniaxial tensile strength σt, the maximum tangential stress of the surrounding cave σθ, the elastic deformation index Wet and the integrality coefficient of rock KV are selected as evaluation indexes. Then taking several rockburst instance data at home and abroad as samples, the rockburst prediction accuracy of XGBoost algorithm is calculated through multiple cross validation, and comparing with the accuracy obtained by support vector machine algorithm and random forest algorithm. Finally, the importance of evaluation indexes is analyzed. The results show that: (1) When with a small number of samples, the randomness of division and sorting for samples has great influence on prediction results, and the reliability of the results can be improved by calculating the average values of the prediction results through multiple cross validation. (2) Among the evaluation indexes, KV and σθ are the most important, while σc is the least important. (3) Due to high prediction accuracy, XGBoost algorithm has some applicability in the field of rockburst intensity classification prediction.

Key words: rockburst prediction, cross validation, XGBoost, reliability analysis, index importance