Crash risk analysis for the mixed traffic flow with human-driven and connected and autonomous vehicles

In the near future, traditional or low-automation vehicles will share the roads with Connected and Autonomous Vehicles (CAVs) over many years. Yet, this complexity may impose new unknowns on the real-time crash risk evaluation. Consequently, it is important to explore crash risk analysis in such kind of mixed traffic flow environments. This paper constructed several special traffic variables in mixed traffic flow environments and proposed the kernel logistic regression (KLR) model to evaluate the crash risk in real-time. A simulated urban expressway corridor based on the North-South Elevated Road in Shanghai, China, is developed in SUMO, for the purpose of collecting the traffic safety data and traffic data (i.e., virtual detector data and Global Navigation Satellite System (GNSS) data) in mixed traffic flow environments. The prediction performance of KLR models was tested and analyzed with the simulated data, and is also compared with that of support vector machines (SVM) models. The results show that KLR has a good prediction performance like SVM. Considering KLR can provide probability estimates directly and can naturally extend to multi-class classification, priority should be given to KLR in similar problems, especially when crash risk is classified into multiple levels. The proposed KLR model is therefore recommended and has the potential to evaluate the real-time crash risk in the mixed traffic flow environment.