Do Drivers' Behaviors Reflect Their Past Driving Histories? - Large Scale Examination of Vehicle Recorder Data

Abstract

We present a method for analyzing the relationships between driver characteristics and driving behaviors on the basis of large-scale and long-term vehicle recorder data. Previous studies relied on precise data obtained under critical driving situations, which led to overlooking routine driving behaviors. In contrast, we used a dataset that was sparse but large-scale (over 100 fleet drivers) and long-term (one year's worth) and covering all driving operations. We focused on classifying drivers by their accident history and examined the correlation between having an accident and driving behavior. We were able to reliably predict whether a driver had recently experienced an accident (f-measure > 86 %). This level of performance cannot be achieved using only the drivers' demographic information. We also found that taking into account the driving circumstances improved classification performance and that driving operations at low velocity were more informative. This method can be used, for example, by fleet driver management to classify drivers by their skill level, safety, physical/mental fatigue, aggressiveness, and so on.