Research on Traffic Safety Evaluation Method Based on Aggressive Driving Behaviors and Traffic Flow Characteristics

Abstract

This study proposes an innovative traffic safety evaluation method based on aggressive driving behaviors (ADBs) and traffic flow from navigation and roadside detector systems to evaluate traffic risks on the National Trunk Highway System (NTHS). We present a traffic risk score (TRS) method to minimize the restrictions of crash occurrence or crash-related data in traffic risk assessment. The non-negative matrix factorization (NMF) algorithm is applied to extract latent risk factors and compute the TRS as it can more effectively deal with comprehensive decision-making problems under multiple factors and is more interpretable. Risk identification ability and interpretability are then demonstrated to measure the performance of the evaluation model. The results show that the aggregated risk area (ARA) is defined and located, conforming to approximate Gaussian distribution in high-risk road frequency space distribution. An NMF-based method can better discriminate high and low-risks roads than other matrix factorization method. A nonlinear relationship between the TRS and the number of accidents is discovered, and roads with more accidents are associated with a higher TRS. Additionally, a more reasonable and interpretable loadings matrix is obtained by the non-negative constraint, which represents the importance of risk factors on the latent risk factor. The method proposed in this paper provides fresh perspectives and suggestions for the discovery of safe-affecting factors and proactive prevention and can assist the traffic management department in alleviating macro-level traffic risks.

Keywords: traffic safety assessment, traffic safety evaluation method, the National Trunk Highway System, non-negative matrix factorization, performance evaluation