Categorizing Driving Patterns based on Telematics Data Using Supervised and Unsupervised Learning

摘要

Progressive motor insurance companies use telematics, a way to quantify the amount of risk involved directly on the usage of the vehicle. Telematics for motor vehicles is a device that records all driving data in real-time, including timestamp based acceleration, GPS location, etc. Current research deals with either only telematics data or insurance claim history to determine a suitable insurance policy. A combination of both the aforementioned data enabled us to create an adaptive method is proposed wherein the categorization of driving patterns would always cover the entire spectrum of driving quality, no matter how good or bad every one of the customers aggregately becomes. This ensures equal penalty and reward when aggregated on the entire customer base which amounts to a stable low-risk policy for insurers while being justifiably attractive to customers. Our technique involves using both supervised and unsupervised Machine Learning algorithms in a two-layer approach to calculate an adaptive driving score for each motor vehicle over some time. This score is nothing but a value indicative of how good a motor vehicle under consideration is driven. A comprehensive explanation is presented about our methodology backed by a detailed evaluation of our two-layered approach and discuss the combination of traditional policymaking and telematics on providing a new-age personalized motor insurance policy for insurers.