Diagnosis and Prediction of Traffic Congestion on Urban Road Networks Using Bayesian Networks

This paper proposes a Bayesian network (BN) analysis approach to modeling the probabilistic dependency structure of causes of congestion on a particular road segment and analyzing the probability of traffic congestion given various roadway condition scenarios. A BN approach was used to encode the joint probability distribution over a set of random variables that described scenario variables, which represented factors affecting the congestion level of a target segment such as time of day, incident, weather, and traffic states on adjacent links, as well as output variables, which represented traffic performance measures of the target segment such as flow, density, and speed. The study developed a method to build a BN model according to historical traffic and event data and demonstrated the BN-based traffic analysis with a study network in Brisbane, Queensland, Australia. The paper discusses applications of the proposed BN model in urban traffic congestion management, by focusing on identifying leading causes for congestion diagnosis and identifying critical scenarios for congestion prediction.