Machine Learning Model for Predicting the Height of the Water-Conducting Fracture Zone Considering the Influence of Key Stratum and Dip Mining Intensity

Predicting the height of the water-conducting fracture zone (WCFZ) is crucial for preventing water inrush and ensuring safe underground mining operations. In this study, we propose a novel model combining CatBoost, XGBoost, and AdaBoost with SSA, HHO, and LEA. Key stratum data (DK, TK) and dip mining intensity data were integrated into the existing parameters for WCFZ height prediction. The main influence angle tangent, derived from the probability integral method, replaces the hard rock ratio coefficient. A total of 104 field datasets with eight input parameters were used, with WCFZ height as the dependent variable. The model was validated using five-fold cross-validation and evaluated with root mean square error (RMSE), mean absolute error (MAE), R2, and mean relative error (MRE). The Preference Ranking Organization Method for Enrichment Evaluations (PROMETHEE) was applied to rank the models. The CAT-HHO model demonstrated the best performance. Using this model, predictions of WCFZ height under varying dip mining intensities showed an approximately linear relationship. SHAP analysis identified mining thickness as the most influential factor. Removing key stratum data from models significantly reduced prediction accuracy. The results highlight the model’s ability to improve WCFZ height prediction, offering insights for water inrush prevention in coal mining operations and providing guidance for applying machine learning to similar challenges.