Modeling the Dynamics of Driver's Dilemma Zone Perception Using Machine Learning Methods for Safer Intersection Control

The "dilemma zone" (DZ) is defined as the area where drivers approaching a signalized intersection must decide to either proceed or stop at the onset of the yellow indication. Drivers that might perceive themselves to be too close to an intersection for a safe stop, and too far to proceed without violating traffic regulations, are said to be caught in DZ. Despite the vast body of related literature, there is a critical gap in research related to the "dynamic nature of drivers' decision" in dilemma zones. In order to identify and capture all significant factors beyond existing research, a driver survey was administered in the three states of Virginia, Pennsylvania, and Maryland. State-of-the-art techniques in human psychology, experimental design, and statistical analysis were used to design the survey and interpret the results. A driving simulator study was conducted to investigate the dynamic nature of driver perception of the dilemma zone and to assess significant factors affecting a driver's decision at the onset of yellow. In addition, the use of machine learning methods to capture the effect of a driver's learning/dynamic perception of DZ was investigated. Findings from this research suggest that drivers do learn from their experience and also that agent-based models can be used for modeling driver behavior in the dilemma zone more accurately than models that currently exist in the literature. The research team therefore recommends that agent-based modeling and simulation techniques should be used for assessing the impacts of dilemma zone mitigation strategies.