Lane-changing behavior is one of the basic driving behaviors and a main contributor to rear-end and sideswipe collisions. Efforts have been devoted to mitigating the hazard risk attributed to lane-changing behavior, by reducing driver’s workload and eliminating human error in the lane-changing process. To this end, it is crucial to have robust prediction of driver’s lane-changing intention for the development of automated driving technology and advanced driver assistance system. Taking the advantage of advanced video image recognition and traffic detection technologies, it is possible to capture comprehensive environment, traffic and driver data for the calibration of prediction model for the lane-changing intention. However, it is not yet known that what features are contributing more to driver’s lane-changing intention. In this study, a meta-analysis was conducted to identify the optimal set of environment, traffic and driver variables that can maximize the efficiency and predictive performance of lane-changing intention, based on 26 selected studies, with which the publication bias was accounted using funnel plot and trim-and-fill methods. In addition, random effect regression model was adopted to measure the association between lane-changing intention and possible influencing factors, controlling for the effects of unobserved heterogeneity. Results of meta-analysis indicate that traffic characteristics (β1=0.42) are more correlated to lane-changing intention, compared to environmental ((β2=0.38) and driver ((β3=0.39) variables, even that the differences are incremental. Furthermore, driver variables were not fully utilized in the prediction of lane-changing intention in previous studies. Overall, findings should be indicative to the development of robust automated driving technology for the vehicle manufacturers and government agencies.

Meta-analysis; automated driving; feature selection; lane-changing intention