Modeling The Frequency of Traffic Conflicts at Signalized Intersections using Generalized LinearRegression Models

摘要

The primary objective of this study was to identify the potential of using conflict prediction models topredict the frequency of traffic conflicts at signalized intersections. The opposing left-turn conflictswere selected for the development of conflict prediction models. Using data collected at thirtyapproaches at twenty signalized intersections where the permitted left-turn phases were used, theunderlying distributions of the conflict frequency for different volume regimes in different time intervalswere examined. It was found that the conflict frequency generally followed a negative binominaldistribution. Different conflict prediction models were developed, including a linear regression model,an overall negative binomial model, and separate models developed for four traffic scenarios whichwere defined based on the volume to capacity ratio of the conflicting traffic flows. The predictionperformance of different models was compared. It was found that the linear regression model was notappropriate for modeling the conflict frequency data. In addition, drivers behaved differently underdifferent traffic conditions. Thus, the effects of conflicting traffic volumes on conflict frequency weredifferent in different traffic conditions. The generalized linear regression models developed for differenttraffic scenarios provided the best estimates for the field measured conflicts.