Abstract: In order to improve the prediction and early warning of road geological hazards and mitigate the impact of heavy rainfall on the safety of high-speed driving in mountainous areas, this paper combines precipitation data from national meteorological stations with data from traffic meteorological stations along highways in Jiangxi Province. Based on the analysis of the geological environment conditions and rainfall characteristics along the highways, four machine learning methods including Support Vector Machine (SVM), logical regression, K neighbors and random forest were adopted to do research on the highway geological disaster forecast modeling and early warning test. The results show that: (1) The majority of geological disasters along Jiangxi highways are located at altitudes of 300 to 450 meters, with slope gradients mostly ranging from 20° to 35°. As terrain slope increases, a unimodal distribution of hazards is observed. Regions with dense river networks and certain vegetation coverage are more prone to experiencing geological hazards. (2) Three main types of rainfall inducing highway geological hazards are identified: long-term rainfall, short-term rainfall, and short-time rainfall. (3) Comparative assessment of four kinds of geological hazard machine learning methods dedicated to geological disasters demonstrates that, for rainfall-induced geological hazards, all four predictive models achieve accuracies exceeding 0.75. Further study found that the logistic regression and random forest model outperform others in forecasting accuracy for both long and short rainfall periods, while the K-neighbor approach was better for short-term rainfall forecast.