To enhance the accuracy and interpretability of tunnel rockburst risk assessment, 386 sets of rockburst case datasets is collected, an XGBoost rockburst risk prediction model integrated with the SHAP algorithm is developed, and the impact of eight different combinations of rockburst risk assessment indicators on model performance is analyzed. The results indicate that: The feature set including all input indicators achieves the optimal performance in predicting rockburst risk levels. Compared with other machine learning techniques, the XGBoost algorithm demonstrates superior applicability in addressing the multi-input single-output challenges of rockburst risk prediction, owing to the advantages of its ensemble learning framework. Additionally, the SHAP algorithm unveiled that the elastic energy index and tunnel depth exert significant influences on the prediction outcomes. These findings offer can provide critical guidance for engineering practitioners and on-site decision-makers to mitigate potential rockburst hazards and formulate science-based mitigation strategies.