Objective: Monitoring and providing early warnings of spontaneous combustion fires in coal pose a significant challenge, hindering the safe advancement of the coal industry. Currently, the accuracy of single-indicator analysis in existing monitoring and early warning methods is insufficient. Multi-indicator judgment methods that rely on statistical analysis are limited by the number and types of indicators, making the judgment process complex and resulting in significant differences. Support vector machine algorithms that are capable of learning rules from limited samples, show potential for application in coal spontaneous combustion monitoring and early warning. Methods: This study establishes a testing system for infrasound waves and acoustic emission signals during coal spontaneous combustion. It explores the relationship between the main frequency amplitude of these signals and temperature to determine whether they can serve as feature vectors for support vector machines. Based on this relationship, the coal spontaneous combustion process is divided into three stages: early, middle, and late, and a combustion support vector machine model is established. The models are trained with different kernel functions, and the one with the highest recognition accuracy for the three periods early, middle, and late stages of coal spontaneous combustion for further validation. Finally, untreated experimental data is used to validate the model's recognition performance. Results: (1) There is a positive correlation between the amplitude of infrasound waves and the main frequency of acoustic emission with temperature, and the correlation coefficient R2 is high, all of which are above 0.90. This shows their effectiveness as indicators for monitoring coal spontaneous combustion. (2) The subsonic polynomial kernel support vector machine can accurately identify the three periods before, during, and after coal spontaneous combustion, outperforming linear and Gaussian kernel support vector machines. Meanwhile, the acoustic emission Gaussian kernel support vector machine surpasses the polynomial and linear kernel models in accuracy for the same phases. (3) The infrasonic support vector machine achieves classification accuracies of 97.75% for the early stage, 97.60% for the middle stage, and 100% for the late stage of coal spontaneous combustion. The acoustic emission model reaches accuracies of 95.65% for the early stage, 95.20% for the middle stage, and 90.20% for the late stage. Conclusions: The multiclass support vector machine model for coal spontaneous combustion presented in this study can accurately identify and classify the state of coal spontaneous combustion. It holds practical significance in coal spontaneous combustion monitoring and early warning. This study introduces a novel method for efficient monitoring and early warning of coal spontaneous combustion.