Risk modeling in smart insurance based on telematics data

The article proposes a model for the use of telematic devices that collect online information about the features of the movement of a vehicle in motor insurance. In the process of developing models using these data, a comparative analysis of different types of devices was carried out, which allows collecting information about the position, speed, accelerations made by the car while moving, and forming an idea of the driving style of its driver. The advantages and disadvantages of the main types of these devices, the features of data storage and collection are described, and the most effective ones from the point of view of insurance tasks are identified. The formats of data coming from telematic devices are described and mechanisms for their aggregation into a convenient, from the point of view of further modeling, information array are proposed. An empirical study was conducted using high-frequency telematics data processed using statistical methods and used to build econometric models. A typification of accidents, in which the drivers included in the study sample fell into, is proposed. Based on the available information about driving conditions, all observations are divided into multiple clusters. Using all available information, risk assessment models were built for different types of accidents for different clusters of observations, and for each of them an optimal set of factors determining the level of accidents was found. An assessment of the quality of the models is made and it is shown to what extent the quality of the models improves using acceleration data. The paper shows how, based on the accident probability assessment model, a service can be formed to prepare driving style recommendations for clients of insurance companies.