Model-Based Semisupervised Clustering

• Activity Number: 284
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Volodymyr Melnykov and Wei-Chen Chen* and Ranjan Maitra+
• Companies: North Dakota State University and Iowa State University and Iowa State University
• Keywords: semi-supervised clustering ; mixture model ; EM algorithm ; gene expression ; microarray data

Abstract

Semi-supervised clustering groups observations in a scenario where only some group identifies are available. We provide a model-based approach to this problem. Model parameters are estimated by the expectation-maximization (EM) algorithm, for which initialization strategies are also developed. A rigorous significance-based approach to estimating number of components is established and has better performance than other information criteria. Simulation experiments in a wide range of settings show improvements in predictions of number of components and classification. The method is applied to finding co-regulated expressed genes in a microarray gene expression study.