Abstract: In order to realize rapid and accurate warning of tunneling faces gas outburst risk, based on the relationship between the characteristics of gas emission in working face and the “three elements” of outburst, a real-time early warning index system including in-situ stress coefficient, gas concentration and gas emission coefficient was established. We combine the SVM and PSO algorithms to build the PSO-SVM outburst warning model, defines the classification principles of outburst warning labels. On this basis,?an outburst warning system of heading face is developed by integrating Spark big data platform. The system includes 8 modules, such as model management, risk identification and Spark configuration. Taking the monitoring and control system of heading face in a mine in Guizhou as the data source, 1059 groups of early-warning indicators and corresponding early-warning grade labels were selected and imported into the data mining model for intelligent learning and training, and the system was applied to the outburst risk early-warning of the heading face. The operation results show that the prediction accuracy of the outburst warning model test set is 92%, and the system can accurately predict the outburst dynamic phenomenon 22 min before the occurrence of the working face.