Trajectory Data Based Clustering and Feature Analysis of Vehicle Lane-Changing Behavior

Abstract

Microcosmic driving behavior includes car-following and lane-changing. Lane-changing may induce traffic oscillations and incidents. In real traffic circumstances, each driver has different attributes, resulting in different lane-changing behavior. However, most studies often neglect the driver heterogeneity due to the complexity of lane-changing modeling. As a result, accuracy and reliability of the lane-changing models were restricted to a certain extent. To address this issue, based on the lane-changing trajectory data extracted from NGSIM, this paper smooths the data by regression. Then, the AP clustering algorithm is applied to classify the drivers based on the pre-processed lane-changing trajectory data. The lane-changing attributes of different types of drivers are compared and analyzed.