Author Description

The presence of adverse weather has a significant negative impact on driving. This research investigated driver lateral behavior under adverse weather via Big Data analytics, Machine Learning, Data Mining in addition to traditional parametric modeling using trajectory-level SHRP2 Naturalistic Driving Study datasets. Initially, driver lane-keeping behavior in adverse weather was examined using ordered logistic regression approach, which indicated that environmental, traffic, driver, and roadway characteristics affect lane-keeping ability. The following study leveraged association rules mining that demonstrated a high association of affected visibility with poor lane-keeping performance. This research was then extended to investigate lane-changing characteristics, which revealed that conservative drivers had longer lane-changing durations in heavy fog compared to clear weather. Moreover, the research provided extensive evaluation into another lateral behavior, named lane-changing gap acceptance, using Multivariate Adaptive Regression Splines. The findings illustrated that relative speed between lane-changing and lead vehicle, acceleration of lane-changing and following vehicle, traffic conditions, and roadway geometries have effects on gap acceptance behavior. Subsequently, emphasis has been provided on developing reliable, accurate, and efficient Machine Learning-based lane change detection and prediction models through a data fusion approach considering different data availability. Finally, the research focused on developing weather-based microsimulation lane change models indicating that weather-specific lane changes were unique and hence, microsimulation models should be weather-specific. The outcomes of this research have significant implications, which could be used in microsimulation model calibration related to lateral behavior and safety improvements in Connected and Autonomous Vehicles, especially in adverse weather.