During deep tunnelling using drill-and-blast method，excavation damaged zone （EDZ） is inevitably induced in surrounding rocks due to the coupled impacts of blast loading and dynamic initial stress unloading and thus affect the structure stability.Therefore，it is very important to predict EDZ depth before roadways excavation.Relying on the field measurements of EDZ in several underground mines as the research object，300 data samples were collected.Four mainstream hyperparametric optimization algorithms，i.e.，genetic algorithm （GA），gray wolf optimization algorithm（GWO），particle swarm optimization algorithm（PSO），and salp swarm algorithm （SSA），were used to optimize the XGBoost algorithm and to construct four hybrid models for EDZ prediction.Comparative analysis of predictive model performance was conducted in terms of R²，RMSE，MAE and MAPE，along with a sensitivity analysis of the influencing parameters.Finally，the optimal PSO-XGBoost model was applied to a transportation roadway in an underground mine for engineering validation.The results show that the GA-XGBoost，GWO-XGBoost，PSO-XGBoost，and SSA-XGBoost models achieve the best predictive performance with swarm sizes of 90，70，60 and 100，respectively.Among them，the PSO-XGBoost model demonstrates the best predictive performance with correlation coefficients of 0.9244 and 0.8787 in the training and testing sets，respectively.Moreover，compared to bench models（XGBoost，RF，SVM and LightGBM），both the prediction accuracy and stability of the optimized models are improved.The tunnel diameter（TD） and rock mass geological strength index（GSI） have the most significant influence on the loosened zone thickness，along with a noticeable impact from the vertical principal stress.The application results of the optimized XGBoost model in practical engineering show that the error between the measured value and the predicted value is within 10%，indicating that the PO-XGBoost is of significance for engineering application.