Rockburst has a large impact on the safe mining of underground mines, and accurate and reliable prediction of rockburst is of great significance to safe production. In order to accurately and effectively predict rockburst, firstly, the stress coefficient, brittleness coefficient, elastic energy index, and integrity coefficient are selected as prediction indicators, and the rockburst prediction index system is established. Secondly, based on the hierarchical analysis method (AHP), improved CRITIC method, and the distance function discriminant method for determining the integrated constant weight, the constant weight is dynamically corrected using the variable weight theory. Then, the effectiveness is introduced to modify the principle of maximum posterior probability, and a Bayesian rockburst intensity grade prediction model with variable weight is constructed. Finally, 15 groups of rockburst engineering examples are selected to verify the reliability and compare the accuracy of the model, and the model is used to predict the rockburst in the middle section of the 1000 m level of Jinchuan No. 2 Mine. The research results show that the accuracy of the Bayesian rockburst prediction model with variable weight can reach 93%. The rockburst intensity grade in the middle section of the 1000 m level of Jinchuan No. 2 Mine is none to slight, and the harm of rockburst is relatively small. The prediction results are in line with the actual situation, which can provide a basis for rockburst prevention and control.

Rockburst is a common ground pressure hazard in underground geotechnical engineering. To predict rockburst accurately in real-time, this study proposes a short-term rockburst prediction model based on microseismic monitoring and probability optimization naive Bayes. Firstly, based on 114 sets of rockburst sample data, four microseismic parameters were selected as predictors using the correlation feature selection algorithm: cumulative number of microseismic events, cumulative microseismic energy, cumulative microseismic apparent volume, and cumulative microseismic energy rate. Secondly, to weaken the conditional independence assumption of the naive Bayes algorithm, the criteria importance through intercriteria correlation method and the similarity function are used to optimize the conditional probability in terms of both attribute weighting and instance weighting. The Mahalanobis distance is introduced to compensate for the loss of prior probability, addressing the decision imbalance caused by conditional probability weighting. Thus, a probability optimization naive Bayes algorithm with conditional probability weighting and prior probability compensation mechanism is proposed to predict the rockburst intensity levels. Finally, the model’s accuracy and reliability are tested through model evaluation, model comparison, and engineering validation. The results show that the proposed model has a prediction accuracy of 86.96% and outperforms other machine learning models, providing a scientific basis for rockburst prediction in practical engineering.