Cluster analysis and multi-level modeling for evaluating the impact of rain on aggressive lane-changing characteristics utilizing naturalistic driving data

Abstract This study investigated lane-changing characteristics with regard to drivers’ aggressiveness in rain and clear weather utilizing the SHRP2 Naturalistic Driving Study (NDS) dataset. An innovative methodology was developed to identify lane-changing events and extract corresponding parameters from the SHRP2 NDS database. Initially, K-means and K-medoids clustering methods were examined to classify drivers into non-aggressive and aggressive categories considering six features related to driving behavior, and K-means clustering was adopted based on the average silhouette width method (ASWM). Two-level mixed-effects linear regression models were calibrated to assess the contributing factors that affect lane-changing durations, which revealed that different vehicle kinematics, traffic, driver, and roadway characteristics, as well as weather conditions combined with other factors, were significant in the calibrated models for both driver types. The results revealed that the lane-changing duration associated with heavy rain decreased with a higher speed limit for aggressive drivers. Furthermore, the lane-changing duration associated with light/moderate rain decreased with the number of lanes for non-aggressive drivers. The study findings could be leveraged to incorporate drivers’ aggressiveness into microsimulation lane-changing model calibration and validation as well as could have significant implications in improving safety in Connected and Autonomous Vehicles (CAV).