In the complex traffic environment, drivers' perception of risk could better predict the potential danger, which helps drivers to take timely measures to avoid the occurrence of collision accidents. Therefore, it is helpful to take targeted measures to improve drivers' safety driving ability by accurately identifying drivers' perception of risk ability. To explore the predictors of drivers' perception of risk ability, typical hazardous scenarios were selected and a risk driving behavior test platform was built based on driving simulator. A total of 35 drivers was recruited for driving simulation tests to collect data on drivers' driving behavior in hazardous scenarios. Based on the TTC, drivers' perception of risk could fall into three categories. Based on the Bayesian linear discriminant method, the support vector machine (SVM), and multinomial logit model, the prediction model of driver's perception of risk ability was constructed. The models constructed by different methods to drivers' perception of risk ability were compared based on the precision rate and F1-score. The results show that the support vector machine has the best predictive performance for the perception of risk. Bayesian linear discriminant method is the second, multinomial logit regression is the worst. In future mixed traffic conditions, it is of great significance to predict drivers' perception of risk ability for improving traffic safety.
 

traffic safety;driving simulation;hazard perception;driving behavior;discriminant analysis