Abstract <p>This paper takes the working face of Heiyanquan Mine as the research subject. Based on the mechanism of coal self-ignition and the gas generation pattern, combined with coal oxidation and temperature rise experiments and machine learning algorithms, a prediction and early warning model for spontaneous combustion risk in goaf areas is constructed. The results show that the double-layer random forest model optimized by grid search and cross-validation has a mean absolute error (MAE) of 6.47 and a determination coefficient (R²) of 0.9845, with a prediction relative error between 0 and 5%. When the working face was advancing slowly through a fault, the model projected that the peak temperature at a depth of 72 meters in the goaf area reached 49.2°C and surpassed 40°C for five consecutive instances, indicating a spontaneous combustion risk. After calculation and numerical simulation, a prevention and control plan was determined with a nitrogen injection flow rate of 9 m³/min and a nitrogen injection port 65 meters away from the working face. During the nitrogen injection process, the system continuously predicted and warned. When the measurement point No. 4 was located at a depth of 105 meters in the goaf area, it was determined that there was no risk of spontaneous combustion, and nitrogen injection was stopped. The findings offer a scientific, accurate, and efficient method for controlling spontaneous combustion in coal mine gob areas, which helps enhance the effectiveness of fire prevention and extinguishment.</p>