Multiple-prototype classifier design

Author

Bezdek, James C. ; Reichherzer, Thomas R. ; Lim, Gek Sok ; Attikiouzel, Yianni

Author_Institution

Div. of Comput. Sci., Univ. of West Florida, Pensacola, FL, USA

Volume

28

Issue

1

fYear

1998

fDate

2/1/1998 12:00:00 AM

Firstpage

67

Lastpage

79

Abstract

Five methods that generate multiple prototypes from labeled data are reviewed. Then we introduce a new sixth approach, which is a modification of Chang´s (1974) method. We compare the six methods with two standard classifier designs: the 1-nearest prototype (1-np) and 1-nearest neighbor (1-nn) rules. The standard of comparison is the resubstitution error rate; the data used are the Iris data. Our modified Chang´s method produces the best consistent (zero-error) design. One of the competitive learning models produces the best minimal prototypes design (five prototypes that yield three resubstitution errors)

Keywords

errors; pattern classification; unsupervised learning; 1-nearest neighbor rule; 1-nearest prototype rule; Iris data; competitive learning; consistent design; labeled data; minimal prototypes design; modified Chang method; multiple-prototype classifier design; resubstitution error rate; zero-error design; Computer science; Error analysis; Hypercubes; Iris; Marine vehicles; Nearest neighbor searches; Pattern recognition; Prototypes; Stock markets; Testing;

fLanguage

English

Journal_Title

Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE Transactions on

Publisher

ieee

ISSN

1094-6977

Type

jour

DOI

10.1109/5326.661091

Filename

661091