Machine Learning for Automobile Driver Identification Using Telematics Data

Abstract

Recent years have seen rapid developments in the automotive industry and the Internet of Things (IoT). One such development is the use of onboard telematic devices that generate data about the car and the driver’s behaviour. This data can be used for identifying drivers through their driving habits. This paper proposes a novel driver identification methodology for extracting and learning driving signatures embedded within the telematics data. First, features representatives of driving style are extracted and derived such as longitudinal acceleration, longitudinal jerk and heading speed from the raw telematics data of GPS coordinates, speed and heading angle. Next, statistical feature matrices are obtained for these features using sliding windows. Finally, several traditional machine learning models are trained over these matrices to learn individual drivers. Results show a driver identification accuracy of 90% for a dataset consisting of only two drivers; the accuracy falls gradually as the number of drivers increases.