Evaluating individual risk proneness with vehicle dynamics and self-report data - toward the efficient detection of At-risk drivers.

Vehicle-dynamics data, now more readily available thanks to moderate-cost, embedded data logging solutions, have been used to study drivers' behavior (acceleration, braking, and yaw rate) through naturalistic driving research aimed at detecting critical safety events. In addition, self-reported measures have been developed to describe these events and to assess various individual risk factors such as sensation seeking, lack of experience, anger expression while driving, and sensitivity to distraction. In the present study, we apply both of these methods of gathering driving data in order to assess risk proneness as accurately as possible. Data were obtained from 131 drivers, who filled in an introductory questionnaire pertaining to their driving habits. Their vehicles were equipped with an external, automatic data-capture device for approximately two months. During that period, the participants reported critical safety events that occurred behind the wheel by (a) pressing a button connected to the device and (b) describing the events in logbooks. They also filled in weekly questionnaires, and at the end of the participation period, a final questionnaire with various self-reported measures pertaining to their driving activity. We processed the data by (a) performing a multiple correspondence analysis of the characteristics assessed via the automatic data capture and self-reports, and (b) categorizing the participants via hierarchical clustering of their coordinates on the dimensions obtained from the correspondence analysis. This allowed us to identify a group of drivers (n = 43) at risk, based on several self-reported measures, in particular, their recent crash involvement, and the frequency of critical acceleration/deceleration events as an objective measure. However, the at-risk drivers did not themselves report more critical safety events than the other two groups.