Time-Varying Parameters in Mixture Models

• Activity Number: 311
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Yorghos Tripodis*+ and Kleoniki Vlachou and David J. Lovell
• Companies: Boston University and University of Maryland and University of Maryland
• Keywords: EM algorithm ; mixture models ; Kalman Filter ; Network traffic

Abstract

Mixture models are widely used in many applications. In this paper we consider the case where some of the parameters of the mixture model vary in time. We use EM algorithm to estimate the parameters of the latent distributions at each point in time and unobserved component time series models to model and estimate the temporal path of the parameters. As our motivation, we use data of actual flight times provided by the National Airspace System. Each flight time on a given pair of airports includes the unimpeded flight time and the time of delay which are both unknown. Having good estimates of unimpeded travel times is very important for queuing models which aim to reduce air traffic congestion. Our model provides the expected unimpeded travel time at each time point, as well as the best linear forecasts. We also compare this method, with methods currently used by transportation engineers.