Risk Assessment Model for Highway Accidents Using Bayesian Random-Parameters Regression Model

This paper presents a highway accident risk assessment model based on a Bayesian random-parameters logit model, aiming to evaluate the effects of real-time traffic conditions on crash risks on freeways. By incorporating random parameters to account for variations in the impacts of traffic variables across different freeway segments, the model offers greater flexibility and adaptability compared to traditional fixed-parameters logit models. The study utilizes traffic flow data collected from the Hangzhou-Shanghai-Ningbo expressway over a 14-month period, analyzing factors such as traffic density, average vehicle speed, and lane-changing frequency. The estimation process employs Markov Chain Monte Carlo (MCMC) methods, including Gibbs sampling and Metropolis-Hasting algorithms, to ensure model convergence and stability. Empirical results demonstrate significant impacts of these traffic variables on crash risks and successfully identify key variables with random effects, enhancing the accuracy of crash risk predictions. This study provides a theoretical foundation and technical support for the development of real-time crash warning systems, contributing to improved highway traffic safety management.