The Real Sorting Hat – Identifying Driving and Scanning Strategies in Urban Intersections with Cluster Analysis

Identifying individual driving strategies often relies on theoretical task models, arbitrary group divisions, or somewhat untransparent evaluations by instructors. We propose using cluster analysis as an exploratory, data-driven approach to categorize drivers based on their driving and scanning behavior. Therefore, we analyzed a combination of variables regarding longitudinal vehicle guidance, lateral vehicle guidance, and gaze behavior when approaching an intersection. Data stemmed from a driving simulator study including drivers with normal vision, simulated, and pathological visual field loss. They performed 32 intersections that varied concerning complexity and the availability of an auditory scanning assistant. The total sample comprised 2145 data points. K-means on two dimensions of a prior Principal Component Analysis yielded the best results with two clusters that can be interpreted as high acter and low acter, referring to the extent and earliness of gaze shifts as well as the duration of the intersection approach. These two strategy clusters were rated based on performance criteria to check the effectiveness of these strategies for the different driver groups and situations. While high acters were more frequent under complex conditions, this strategy failed more frequently in these cases. Future developments for this promising approach to cluster strategies in driving-related areas are discussed.