Wednesday, December 7th, 2011
2:00pm - 3:00pm, in Science 2-064

Paul Ossenbruggen

UMass Boston

Understanding Traffic Congestion: Investigating It as an Extreme Event

Abstract: Traffic flow data is difficult to model due to inherentlynoisy nature, changing volatility, high temporal dependence, andshifts, often sudden, between congested and free flow states. In spiteof this, accurate measures of flow are critical for roadway design anddeveloping useful capacity estimates. In this presentation we explorethe use of extreme value theory and functional data analysis asapplied to traffic data, and the appropriateness of this approach.Specifically, we perform a peaks over threshold (POT) technique andfit a generalized Pareto distribution to the associated clustermaxima. Results from this methodology allow for parametric inferenceand capacity identification using functional data analysis. Theseresults may prove useful in explaining the "duality effect" -- thephenomenon that at or near capacity of a roadway may or may nottrigger congestion. Evidence will be presented to justify the use ofthis approach to traffic flows, and areas of ongoing research will beidentified. Comments and criticism are encouraged.