The invention relates to a driver risk assessment model training method and device and a risk assessment method and device. According to the invention, data corresponding to dimensions required by a driver model is obtained from various data resources, wherein the data sources comprise GPS data, vehicle-mounted box data, traffic data, weather data, driver data, map data and insurance accident occurrence data, the data sources are very rich, the data for assessment is more comprehensive based on the data sources, and the driver type and score outputted by a trained machine learning classification model can reflect the level of each type of drivers. In addition, the data sources comprise the insurance accident occurrence data, the driver score is comprehensively calculated by giving a weightto the data of each dimension of the driver model, so that the obtained driver type and score are more accurate, and the vehicle insurance price set according to the model is more accurate.