Abstract: To address the weak interpretability caused by the "black-box" structure of current gas concentration prediction models in the upper corner of fully mechanized mining faces, a gas traceability model based on XGBoost-SHAP was proposed for the upper corner of fully mechanized mining faces. Correlation analysis was conducted on the monitoring data of gas emission from fully mechanized mining faces to select feature variables. An upper corner gas concentration prediction model was constructed based on XGBoost, and the SHAP algorithm was introduced to calculate the contribution of each feature variable to the prediction results, thereby enhancing the model's transparency and providing a global interpretation for the XGBoost model. Finally, the model performance was evaluated using multi-source sensor monitoring data from the field. Case analysis results showed that: ① the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE) of the XGBoost model were 0.93, 0.007, and 0.008, respectively, indicating the highest goodness of fit and the lowest errors compared with random forest (RF), support vector regression (SVR), and gradient boosting decision tree (GBDT). ② The mean relative error of the XGBoost model was 4.478%, demonstrating higher accuracy and better generalization performance compared with the other models. ③ Based on the mean absolute SHAP values of input features, the gas concentration at T1 on the working face had the greatest influence on the gas concentration in the upper corner, followed by the gas concentration in the upper corner extraction pipeline, with the gas content and roof pressure of the mining coal seam following closely. These findings indicate that XGBoost can capture the nonlinear relationships and interactions between variables, and that the SHAP algorithm can provide global interpretability for the XGBoost model.