Clustering and Discriminant Analysis via Mixture Models
Department of Mathematics & Statistics
Seminar Presentation
Friday, March 18, 2016
3:30 P.M.
Huggins Science Hall (rm 141)
Paul McNicholas, Ph.D., P.Stat.
Professor & Canada Research Chair in Computational Statistics
Department of Mathematics & Statistics
McMaster University, Hamilton, Ontario, Canada, L8S 4L8
Clustering and Discriminant Analysis via Mixture Models
The application of mixture models for clustering has grown into an important subfield of classification. First, the definition of a cluster is discussed along with some historical context for model-based clustering. Then, some work to date in this direction is reviewed before more recent work is presented. This includes work on dealing with asymmetric clusters and/or outlying points, as well as a discussion of some computational considerations. Real and simulated data are used for illustration. The focus then shifts to discriminant analysis, or supervised classification. Mixture discriminant analysis uses a mixture of, usually Gaussian, distributions to model each class within the training set; the resulting parameter estimates are then used to classify observations in the test set. An alternative approach is presented where one, more flexible, distribution is used to model each class in the training set; a comparison with mixture discriminant analysis is then presented and discussed.