Construction of a Two-Stage Rockburst Warning Model Based on Multi-Source Rockburst Case Studies

Rock burst is a sudden disaster that influenced by various factors. Accurately identifying and predicting rock risks of great significance for improving mine safety. In order to improve the accuracy prediction from perspective data structure, this paper constructs two-level model based on six features. The first-level uses Box-Cox, Yeo-Johnson, uniform transformations scaling extends using CTGAN architecture, followed feature dimension optimization. second-level K-Means algorithm reconstruct labels enhance inter-class differences, visualizes clustering effects ISOMAP. At optimization level, an ensemble stacked with 8 algorithms deep forest are used prediction. results show transformation, increasing complexity, appropriate expansion can effectively accuracy. single achieved maximum 81.25%, established outperformed machine learning method, improvement 17.3%. Feature had highest 6.3%. Through comparison, it was found has 98.6%, which superior other models such as Gradient Boosting Multilayer Perceptron. addition, SHAP value 7 evaluation indicators were evaluate further explain results. proposed provides certain reference accurately bursts.