Modeling Cluster Mobility: A DPM Model for Longitudinal Data

• Activity Number: 579
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Yuda Zhu*+ and Robert E Weiss
• Companies: University of California at Los Angeles Fielding School of Public Health and University of California at Los Angeles Fielding School of Public Health
• Keywords: Nonparametric Bayesian ; Dirichlet Process ; Clustering ; incidence

Abstract

The Dirichlet process mixture (DPM) model has been used extensively as a non-parametric Bayesian model and can be used in clustering problems when the number of clusters are not known a priori. The standard DPM model assumes exchangeable observations from a single unknown distribution. We present the cluster memory Dirchlet process mixture model (cmDPM), an extension of the standard DPM model for the modeling of cluster dependence over time in the nonparametric Bayesian framework. The cmDPM model introduces dependence between distributions over time by retaining observation specific memory of the previous clustering structure. The model is applied to tuberculosis incidence data over the last 20 years and the corresponding findings are presented.