University of Houston
Department of Computer Science
In
partial fulfillment of the Requirements for the Degree of
Master of
Science
Murali krishna Achari
will defend his
thesis
On using Clustering to Enhance Classification
Abstract
In general, we are interested in exploring how class density information obtained though clustering can be used to enhance the classification process without changing the mechanism of existing learning models.
In this thesis we propose a preprocessing step to classification that applies a clustering algorithm to the training set to discover local pattern in the input space. We demonstrate how this knowledge can be exploited to enhance the predictive accuracy of simple classifiers which are characterized by high bias and low variance. These classifiers are limited in their flexibility of decision surfaces and experience difficulty in delineating class boundaries over the input space when a class distributes in complex ways. One way to solve this is to increase the complexity of the classifiers which in turn increases the flexibility of the decision surface. But this comes at an expense of increased variance. Decomposing a class into clusters transforms the input space in such way that makes it easier for the classifiers to approximate the target concept and also provides a viable way to reduce bias without increasing the variance.
Date: Thursday, August 19, 2004
Time: 10:00 AM
Place:
550-PGH
Faculty, students, and the general public are
invited.
Thesis Advisor: Dr. Ricardo Vilalta