Modeling Highway Traffic Volumes.pdf

Modeling Highway Traffic Volumes Toma?s? S?ingliar1 and Milos? Hauskrecht1 Computer Science Dept, University of Pittsburgh, Pittsburgh, PA 15260 {tomas, milos}@cs.pitt.edu Abstract. Most traffic management and optimization tasks, such as accident detection or optimal vehicle routing, require an ability to ade- quately model, reason about and predict irregular and stochastic behav- ior. Our goal is to create a probabilistic model of traffic flows on highway networks that is realistic from the point of applications and at the same time supports efficient learning and inference. We study several multi- variate probabilistic models and analyze their respective strengths. To balance accuracy and efficiency, we propose a novel learning model, mix- ture of Gaussian trees, and show its advantages in learning and inference. All models are evaluated on real-world traffic flow data from highways of the Pittsburgh area. 1 Introduction The importance of road transportation systems to our daily lives can hardly be overstated. To facilitate monitoring and management of their complexities, sensors collecting traffic information (such as traffic volumes and speeds) are installed on many roads. Today, coverage is good for major highways in many metropolitan areas and traffic sensor deployment is rapidly increasing worldwide. Road networks exhibit strong interaction patterns among traffic variables and traffic is subject to stochastic fluctuations. The ability to adequately model, reason about and predict stochastic behavior is crucial to computational support of many traffic management tasks, such as traffic routing, congestion analysis and accident detection. The objective of our work is to develop models of large multivariate continuous probability distributions describing vehicular traffic. We require that the models be compactly parameterized and admit efficient inference and learning algorithms. The quantities of primary interest in this paper are traffic flows. Traffic flow is