Joint Clustering and Registration for Functional Data

• Activity Number: 221
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Yafeng Zhang*+ and Donatello Telesca and Steve Horvath
• Companies: Amgen and University of California, Los Angeles and University of California, Los Angeles
• Keywords: Curve registration ; Curve clustering ; Functional data ; Dirichlet process ; Nonparametric Bayesian statistics ; MCMC

Abstract

Methods have been proposed for registration and clustering for functional data separately. However, those models are not adequate in some situations where simultaneous curve registration and clustering are necessary. In this article, we propose a Bayesian hierarchical model for joint curve registration and clustering. With B-spline representations of shape functions and time transformation functions, our model provides flexible nonlinear modeling for both functions; With a Dirichlet process model for clustering, it identifies both cluster numbers and clustering structure among curves. We test the joint model on simulated datasets and show that it performs better than models for clustering only or registration only. We also apply the model to the Berkeley Growth Data and time course expression data of the response of Human fibroblasts to serum stimulation.