With the rapid advance of the Internet of Vehicles (IOV), IOV data are becoming increasingly available. This kind of data includes plentiful information, which can reflect driving behavior and driving environment. In this work, we try to utilize IOV data for accident prediction. Specifically, we propose a Usage Based Insurance (UBI) model where the response is whether the vehicle is involved in an accident. Mileage, driving behavior as well as Point of Interest (POI) information are incorporated as predictive variables. The newly proposed UBI model can be further used for driver segmentation.