Bayesian Mixture Model--Based Clustering with Unknown Number of Components

• Activity Number: 187
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Taiyeong Lee*+ and Jaeseok Kim and Yongdai Kim
• Companies: SAS Institute Inc. and Samsung Electronics and Seoul National University
• Keywords: Bayesian factor analysis ; MCMC ; Factor model ; mixture model ; Chinese restaurant process

Abstract

We present a clustering algorithm using Bayesian multivariate mixture model for high dimensional data. In particular, we use a Gaussian mixture model in which the covariance matrices are parameterized by factor models in order to reduce computational complexity as well as number of parameters. We propose a modified Chinese restaurant process which allows some factors to be shared by clusters. That further reduces the number of parameters and improves the model interpretability. We also show how the procedure automatically determines the number of components through a reversible jump MCMC algorithm. The performance of new algorithm is illustrated with both simulated and real data examples.