Different types of driver workload are suggested to impact driving performance. Operating a vehicle in a situation where the driver feel uneasy is one example of driver workload. In this study, passenger car driving data collected with Naturalistic Driving Study (NDS) data acquisition equipment was analyzed, aiming to identify situations corresponding to a high driver's subjective rating of ‘unease’. Data from an experimental study with subjects driving a passenger car in normal traffic was used. Situations were rated by the subjects according to experienced ‘unease’, and the Controller Area Network (CAN) data from the vehicle was used to describe the driving conditions and identify driving patterns corresponding to the situations rated as ‘uneasy’. These driving patterns were matched with the data in a NDS database and the method was validated using video data.
Two data mining approaches were applied. The first was based on an ensemble classifier on general variables derived from the CAN-data to predict the subjective rating of segments of the data. The second used hierarchical clustering with a distance metric based on the principal variance components over segments. The ensemble classifier explained a large proportion of the variance when adjusting for driver and route. The hierarchical clustering method performed well, distinct clusters corresponding to a high driver subjective rating could be obtained.
Identifying situations with increased driver workload in NDS data is a complex task addressing a large variation of traffic situations and driver experiences. The proposed method is a first approach to help address this topic using data mining.