Abstract: The accurate evaluation of coal seam gas content is very important, and the prediction accuracy of gas content model is gradually improved with the process of coalbed methane exploration and development. Based on the logging response characteristics of coal seam gas content in Yanchuannan gas field, the MIV(Mean Impact Value) method is used to optimize the logging parameters, and the BP neural network and random forest ideas are introduced to establish the gas content prediction model of coal reservoir respectively. The established model is compared with the traditional multiple linear regression method, and the gas content distribution in the study area is briefly described based on the random forest model. The results show that the simple multiple linear regression model has poor prediction results, and it is difficult to reflect the complex relationship between logging parameters and gas content of coal reservoirs. The prediction accuracy of BP neural network model and random forest model has been significantly improved. Among them, the random forest model has higher prediction accuracy and is more suitable for the prediction of gas content in coal reservoirs in the study area.Based on the prediction of random forest model, the distribution range of gas content in gas field is 4.84-21.83 m³/t, with an average of 11.63 m³/t. On the plane, the gas content of coal seam increases gradually from southeast to northwest, and its variation law is generally consistent with the law of buried depth of coal seam in gas field. Vertically, with the increase of buried depth, the gas content of coal increases gradually, and with the increase of buried depth of coal seam, the dispersion degree of gas content distribution increases.