Coal and Gas Outburst Behavior in Gas‐Bearing Tunnels Revealed Through an Interpretable ISSA‐DNN Learning Framework

Coal and gas outbursts represent one of the most severe safety hazards during the construction of gas‐bearing tunnels. Accurate risk prediction remains difficult because of complex geological conditions, limited sample sizes, and the inherent subjectivity in hazard classification. In this study, an interpretable data‐driven prediction framework is developed by coupling an Improved Sparrow Search Algorithm (ISSA) with a Deep Neural Network (DNN). A dataset comprising 198 samples collected from multiple gas‐bearing tunnel projects is established and carefully preprocessed, including missing‐value imputation, outlier treatment, and class balancing. To reduce subjectivity in historical labels, Principal Component Analysis combined with K‐Means++ clustering is employed to reassign outburst risk levels. The ISSA introduces multiple enhancement strategies to improve global search capability and convergence stability and is used to optimize key hyperparameters of the DNN. Model interpretability is further enhanced using SHapley Additive exPlanations (SHAP), which quantify feature contributions at both global and individual‐sample levels. The proposed ISSA‐DNN model achieves an accuracy of 0.91 and an F 1 ‐score of 0.92, outperforming several commonly used machine learning methods. SHAP analysis indicates that the coal firmness coefficient, initial gas desorption rate, and burial depth play dominant roles in outburst risk prediction. The proposed framework provides a practical and transparent tool for early hazard identification in gas‐bearing tunnel engineering.