Efficient Clustering Algorithms Via Multivariate Techniques and Mixture Models

• Activity Number: 211
• Type: Contributed
• Date/Time: Tuesday, August 13, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing*
• Author(s): Yun-Fei Chen*+
• Affiliation(s): Eli Lilly and Company
• Keywords: clustering ; principal component ; binary decision tree ; mixture model

Abstract

We develop new efficient clustering algorithms based on multivariate techniques and Gaussian mixtures. For two-class and three-class clustering, we first cluster the data points along the first principal component and then apply linear discriminant analysis to refine the clustering. To extend this strategy to multi-class clustering, a binary decision tree approach and an agglomerative linkage technique are employed. Furthermore, when we use a mixture model on the clustered data, more accurate clustering can be obtained. The EM (expectation-maximization) algorithm is adopted for parameter estimation. Simulations show that our algorithms can achieve more accurate clustering and have much faster speed than existing methods.