Abstract: [Objective] Infrasound is an effective debris flow warning approach. The threshold warning method commonly used in previous studies considered only individual infrasound characteristics, which can result in inaccuracies in debris flow warning. [Methods] Infrasound data collected in Caodianshui village, Beijing, was employed to investigate the differences between infrasound characteristics of debris flow and environment background. Based on random forest algorithm, the infrasound warning model for debris flow was established. [Results] The effective pressure of debris flow infrasound ranges from 0.4 to 1.0Pa. In contrast, the effective pressure of environmental infrasound is mostly lower than 0.1Pa, but it can exceed 0.4Pa due to noise. The energy of noise infrasound concentrated in frequency band below 6Hz. The energy of debris flow infrasound in 6-15Hz is significantly higher than that of noise infrasound. Therefore, comprehensive time-frequency characteristics, especially the energy in 6-15Hz, should be taken into consideration to identify debris flow infrasound. By taking effective infrasound pressure, effective infrasound pressure of 6-15Hz, zero crossing rate, main frequency and amplitude of main frequency as characteristic variables, the debris flow infrasound warning model based on random forest was constructed. The AUC of the model was 0.99. The test data recognition accuracy reached 90%, which is 15% higher than that of the traditional threshold method. [Conclusion] This study demonstrate that the infrasound warning model for debris flow based on random forest algorithm can significantly improve the warning accuracy and is applicable to the typical basin in Beijing mountainous area, which can also provide a reference for the debris flow infrasound warning research in other areas.