Traffic Conflict Prediction for Basic Freeway Segments Considering Different Indicators

Freeways are facilities where it’s likely for severe traffic accidents to happen and thus have great safety concerns. This study predicts the safety of basic freeway segments using traffic conflict techniques. From the highD data set, two different conflict indicators, time to collision (TTC) and post encroachment time (PET), were extracted to measure the traffic conflicts, considering different thresholds. Road- and vehicle-factors which could influence the occurrence of traffic conflicts were considered independent variables. Both Poisson distribution models and Negative Binomial models were developed, and their performances were evaluated according to the goodness of fit and prediction accuracy. The result shows that the Negative Binomial distribution models are better because the data is overdispersed. The prediction accuracy of the models using PET is generally higher than that of the models using TTC. Variables such as lane-changing frequency and standard deviation of space-mean-speed have significant impact on the traffic conflicts.