In partial fulfillment of the Requirements for the Degree
of
Master of Science
Waree
Rinsurongkawong
will defend her thesis
The
Cross-Comparison of Nearest Neighbor,
Neural
Network, and Discriminant Analysis Approaches
Abstract
Data mining has gained significant importance in the last decade. In this thesis three popular approaches, neural network, discriminant analysis, and k-nearest neighbors, are evaluated and compared using a benchmark that involves three classification problems . The commercially available MATLAB neural network toolbox was used to conduct several neural network experiments. The comprehensive statistic software tool SPSS was used for the discriminant analysis approach. As part of this thesis a sequential and parallel k-nearest neighbor algorithm was implemented along with several modifications.
For each data set, the different versions of each approach were tested to find the best parameter settings and options. Moreover, a cross-comparison among the three methods was conducted to find which method will obtain the best classification performance for each data set and to answer the question of whether different approaches are making the same or different kind of classification errors. The findings of this cross-comparison suggest that a better classification accuracy can be obtained by combining classifiers that have been derived using several approaches.
Date: Friday, April 6, 2001
Time: 2:00 PM
Place: 550-PGH
Faculty, students, and the general public are invited.
Thesis Advisor: Dr. Christoph F. Eick