Abstract: To address the issue that traditional coal spontaneous combustion temperature prediction models do not consider multicollinearity between indicator gases and temperature data and have insufficient prediction accuracy, a coal spontaneous combustion temperature prediction model using Support Vector Regression (SVR) with hyperparameters optimized by Bayesian Optimization (BO), abbreviated as BO-SVR, was proposed. A programmed heating experiment of coal spontaneous combustion was conducted to collect and process the generated indicator gas data. Spearman correlation analysis was used to select indicator gases with strong correlation to coal temperature and analyze the multicollinearity among the amounts of the generated indicator gases. Principal component analysis was performed on the selected indicator gases to resolve multicollinearity and reduce dimensionality simultaneously. Five-fold cross-validation was used to divide the training set and test set. The performance of the BO-SVR model was quantitatively evaluated in comparison with SVR, Particle Swarm Optimization SVR (PSO-SVR), and Genetic Algorithm-Optimized SVR (GA-SVR) models using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²). Results showed that the MAE of the BO-SVR model decreased by 74.2%, 36.7%, and 10.2% compared with the other three models, respectively; the RMSE decreased by 71.9%, 33.3%, and 11.4%, respectively; and the R² reached 0.9885, which was higher than other models. Parallel experiments were conducted using bituminous coal samples from Shanxi Coal Import and Export Group Hequ Jiuxian Open-pit Coal Industry Co., Ltd. The results showed that the BO-SVR model had an MAE of 4.9279 ℃, an RMSE of 6.4899 ℃, and an R² of 0.9853 on the new dataset, which was highly consistent with the prediction results of the original dataset. This indicates that the BO-SVR model has good generalization ability, prediction accuracy, and robustness, contributing to improving the accuracy of coal spontaneous combustion temperature prediction.