Psychological Driving Style Estimation from GPS Sensor Data Alone

Automobiles are essential to society, but accidents involving older drivers are increasing. Driving Assistance Systems (DASs) have been developed to prevent such accidents. An essential goal is to build on this by implementing a system that provides adaptive assistance without burdening drivers with different characteristics. Accurately estimating driver characteristics is essential for this purpose. In contrast to studies that use in-vehicle sensor data through a Controller Area Network (CAN)-bus data, which requires additional equipment for data collection and is accessible to only a limited number of people, a more efficient and accessible approach is needed. Previous studies have shown that driving styles can be estimated using easily accessible Global Positioning System (GPS) data alone. We posit that the psychological driving style, which is the attitude and orientation for driving and affects the driver’s behavior, can also be estimated from GPS data. In this study, we focus on developing a Driving Style Questionnaire (DSQ) as an indicator of the psychological driving style recognition model. The experimental results reveal differences in some features of all the GPS data between the high and low groups, which are divided by the median for each DSQ item on the two road types. Additionally, the findings indicate that the model with GPS data achieves the F1-macro greater than the random-assignment baseline on five out of eight of the DSQ items in the majority, and the average macro F1-score is 0.596, a difference of 0.036 from 0.632 for the model with CANbus data. The proposed GPS-based recognition model contributes to the development of personalized assistance systems without the burden of excessive sensor installation costs.