Coalbed methane content is one of the key parameters for evaluating the gas hazard levels in mines. Accurately understanding the coalbed methane content and its influencing factors is of critical importance for effective gas management and disaster prevention in mining operations. This study employs four machine learning algorithms—Elastic Net, Support Vector Regression (SVR), Gradient Boosting Decision Tree (GBDT), and Extreme Gradient Boosting (XGBoost)—to establish a predictive model for coalbed methane content. Additionally, the SHAP model is used to explain the feasibility of the machine learning algorithms, reveal the impact of input variables on output variables, and identify the main influencing factors of coalbed methane content. The results show that the GBDT model achieves the highest fitting accuracy (R2 =0.9956) on the prediction set, demonstrating the best performance. Feature importance analysis of the model and global feature importance analysis using SHAP reveal that the ratio of sandstone to mudstone and mudstone thickness are the most significant factors affecting the prediction of coalbed methane content, followed by volatile matter. In contrast, coal seam thickness, bedrock thickness, and ash yield have the least impact on the prediction. GBDT combined with SHAP algorithm effectively obtained the interaction between various factors in the prediction process of coal seam gas content. There is a clear dependence between different features, with the strongest interaction between sandstone and mudstone being the thickness of mudstone. The strongest interaction with the thickness of mudstone is the volatile matter.