Abstract: In order to solve the problems of unreasonable selection of prediction methods and poor prediction accuracy for gas emission in mining face, with the help of machine learning methods, in this paper, existing data was processed through cross-validation, the support vector and the random forest algorithm were used to construct the prediction model and the prediction performance of the two models before and after dimensionality reduction of the input data were compared and analyzed. The test shows that after cross-validation and data dimensionality reduction, the two prediction models tend to be stable, and the prediction accuracy are greatly improved. The overall prediction performance of the SVR model is slightly better than that of the RFR model; SVR model takes about 0.015 s per round of testing, much lower than the latter one; the average absolute error of the SVR model is about 0.18 m3/min, and the average relative error is about 3.26%, which is relatively better than the other one. The models established in this paper both present good performance, and the SVR model after feature screening by random forest takes less time, has less error and better stability, which effectively realizes the prediction of the gas emission and plays a reference role in simplifying data collection and scientifically formulating gas control measures.