Enhancement of Clustering Results with Variable Selection and Resampling Methods

• Activity Number: 585
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Wenzhu Bi*+ and George C. Tseng and Julie C. Price and Lisa A. Weissfeld
• Companies: University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
• Keywords: clustering ; variable selection ; resampling ; imaging ; feature selection ; PET

Abstract

Clustering can be used to identify biologically distinct subgroups from an n×p dataset without knowledge of the true group membership. Since some variables are irrelevant to clustering and may only introduce noise, variable selection methods have recently been developed to exclude these variables and to yield more reliable and parsimonious clustering results. Recently in 2010, Witten and Tibshirani introduced a general framework for variable selection by applying a Lasso-type penalty and an L2 condition. We propose to combine Witten's method with resampling techniques, such as bootstrapping or leave-one-out resampling. The goal is to alleviate the effects of noise and outliers on the variable selection and clustering results and also to generate confidence intervals for the clustering results. The performance of the proposed method is demonstrated by simulation. We then apply the method to neuroimaging data. The focus is on the analysis of voxel-level data and the identification of a subset of voxels that can be used to classify subjects into groups. We present a PET imaging example using an unspecified radiotracer to identify groups of subjects with varying amounts of tracer.