Abstract: Abstract: In order to accurately and quickly predict the coal spontaneous combustion in mine, a classification early warning method of coal spontaneous combustion combined with Bayesian optimization method and XGBoost model was proposed. Based on the experimental data of coal spontaneous combustion in a mine in Shandong province, the early warning grades of coal spontaneous combustion are firstly classified according to the change trend of index gas curve with temperature. Secondly, the data are divided into training set and test set, and the index gases of O2,CO,C2H4,CO/ΔO2,C2H4/C2H6 are selected as the input of the XGBoost model, and the coal temperature grade is taken as the output of the model. At the same time, bayesian optimization is used to optimize the parameters of XGBoost, such as learning rate, n_estimators, maximum depth. The XGBoost model of coal spontaneous combustion classification early warning based on Bayesian optimization algorithm (Bo-XGBoost) is constructed. Then, the test set data is brought into BO-XGBoost, XGBoost, BO-RF, BO-SVM, BO-KNN and other models for comparative analysis. The results show that the prediction accuracy of BO-XGBoost model is 91%, which increased 3%, 9%, 12% than that of BO-RF, BO-SVM and BO-KNN respectively. Finally, the applicability and stability of the BO-XGBoost Algorithm are further verified by its application in the early warning of spontaneous combustion classification in a mine in Tangshan. The results show that the BO-XGBoost model is more suitable for the early warning of coal spontaneous combustion.