In order to accurately and quickly analyze the spontaneous combustion temperature of coal in goaf and avoid spontaneous combustion fire, the SSA-RF regression analysis model combining SSA and RF algorithm was proposed. Firstly, based on the data obtained from the spontaneous combustion characteristics test in Dongtan coal mine, the regression results of SSA-RF model and RF, back propagation neural network (BPNN), particle swarm optimization algorithm (PSO)-BPNN and SSA-BPNN model were compared and analyzed. Then, the reliability of the SSA-RF model was verified by taking the test data of 1204 coal face in Zhengjia coal mining as an example. Finally, the model was applied to Donggucheng coal mine. The results show that the mean absolute errors (MAE) of SSA-RF, RF, BPNN, PSO-BPNN and SSA-BPNN are 11.203 1, 14.342 0, 19.599 1, 15.530 6 and 14.352 8, respectively. The mean absolute percentage error (MAPE) is 14.89%, 16.91%, 18.55%, 18.43% and 18.11%, respectively. The root mean square errors (RMSE) are 13.761 0, 16.525 0, 20.786 6, 18.022 7 and 17.735 5, respectively. The coefficients of determination (R2) are 0.927 4, 0.882 7, 0.815 3, 0.843 6 and 0.868 8, respectively. All indexes of SSA-RF model are the best, which indicates that it is universal and stable, and it is more suitable for regression analysis of coal spontaneous combustion temperature.