George Mason University
AES/SCS Statistics Colloquium Series
Seminar Announcement



Introduction to Model-Based Clustering

Wendy L. Martinez


Office of Naval Research


ABSTRACT

Model-based clustering takes a finite mixture based approach to the problem of finding groups in a set of data. In finite mixtures, a probability density function is modeled as a weighted sum of component densities. The Expectation-Maximization (EM) algorithm is then used to estimate the parameters of the mixture. With this methodology, we hypothesize that each component density is representative of a cluster or group. Several issues must be addressed when this technique is used. How many clusters or component densities should be used? What is a good starting point for the EM algorithm? What 'form' should the component densities take? How can we assess the results of the clustering? This presentation is tutorial in nature and will show the model-based clustering methodology. The concepts will be demonstrated using a MATLAB Model-Based Clustering Toolbox.


Friday, February 14, 2003
George W. Johnson Center, Assembly Room C
Seminar at 10:45 a.m.
Refreshments at 10:30 a.m.
For the 2003 Spring Seminar Schedule, go to
www.science.gmu.edu/statseminars