Exploring Models for Clustering Data

• Activity Number: 86
• Type: Invited
• Date/Time: Monday, July 30, 2007 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Graphics
• Author(s): Dianne Cook*+
• Companies: Iowa State University
• Keywords: model-based clustering ; dynamic graphics ; self-organizing maps ; model diagnostics ; tours ; EDA

Abstract

Clustering data is so much fun, yet it is one of the murkiest of statistical procedures. The usual suspects measuring the goodness of fit, for example, Bayes Information Criterion for model-based clustering, often yield disappointing results. In this talk we'll describe interactive and dynamic graphics methods that help to investigate cluster analysis models. We'll focus on model-based clustering, and examine the parameterization of the variance-covariance matrix, using tours. And we'll also look at the way self-organizing maps weave their net model into multivariate data.