Abstract: The deep mining coal seams in the XiMeng mining area are affected by complex environmental conditions, such as high in-situ stress, large water inflow, and severe air leakage, which increase the risk of coal spontaneous combustion and make prediction more challenging. Coal samples from Yingpanhao and Shilawusu coal mines within the mining area were selected for temperature-programmed spontaneous combustion experiments to determine characteristic parameters of coal spontaneous combustion under different moisture and sulfur mass fraction conditions. Combined with coal quality parameters from industrial analysis, a prediction database was established, and the Crested Porcupine Optimization algorithm (CPO) was applied to optimize the hyperparameters of the random forest (RF) model, CPO-RF model was established to predict the degree of spontaneous combustion of coal. The results showed that the patterns of gas concentration and oxygen consumption rate during the oxidation heating process were similar in the Yingpanhao and Shilawusu coal samples. CO was identified as the main indicator gas, with an initial appearance temperature of about 30 ℃. The amount of gas produced increased with higher sulfur mass fraction, and initially decreased and then increased with higher moisture mass fraction. The critical temperature of coal spontaneous combustion was determined to be 67.5~70.5 ℃, and the dry cracking temperature was 113.5~115.4 ℃. Through CPO’s efficient global search capabilities, the RF model’s optimal tree depth and tree count were automatically identified, avoiding suboptimal solutions caused by improper settings, thus enhancing the model’s generalization and robustness. The constructed CPO-RF model significantly improved the accuracy of coal spontaneous combustion predictions, with high alignment between predicted and actual temperatures on the test set, a mean absolute error and root mean square deviation of 0.762 ℃ and 1.014, respectively, and a coefficient of determination reaching 0.9994. The predicted results of the CPO-RF model, when compared with the characteristic temperatures of coal spontaneous combustion, can enable efficient discrimination of the risk of coal spontaneous combustion. Based on this, targeted fire prevention and extinguishing methods can be adopted. The research findings provide a reference for the prevention of coal spontaneous combustion in deep mining in mining areas.