Accurate prediction of coal seam gas content is an important link to prevent underground gas disasters. In order to improve the scientificity and accuracy of underground gas content prediction, 41 sets of data from different mining areas are obtained, including gas content, buried depth, coal thickness, moisture, ash and volatile content. Five algorithms of least square support vector machine (LSSVM), deep belief network (DBN), Long short-term memory (LSTM), Elman neural network and adaptive enhancement (Adaboost) are selected, and the optimal base model is the least square support vector machine (LSSVM), adaptive enhancement and deep belief network. Seven gas content prediction models are integrated through the base model, and four models of Stacking- LSSSVM-Adaboost, Adaboost, Stacking-Adaboost-DBN and Stacking- LSSSVm-Adaboost-DBN are optimal models. Four prediction and evaluation indexes, namely, decision coefficient, mean absolute error, root mean square error and mean absolute percentage error, were used to comprehensively evaluate the four selected models, and the models with MAE < 0.2, RMSE < 0.3 and MAPE < 10 were selected as the final prediction models for gas content. The results show that the decision coefficient of the integrated Stacking-LSSVM-Adaboost-DBN model is 0.951, and MAE, RMSE and MAPE are 0.170, 0.204 and 7.412, respectively. The established model has high prediction accuracy and can provide a basis for mine gas disaster prevention.