A Genetic Instance-Based Collaborative Approach for Attribute Weightings

Luciana De Nardin (1), Maria do Carmo Nicoletti (2)

e-mails: luciana@pucpcaldas.br, carmo@dc.ufscar.br

(1) Pontificia Universidade Católica de Minas Gerais - Dept. de Ciência da Computação 37701-355 Poços de Caldas Brasil
(2) Universidade Federal de São Carlos - Dept. de Ciência da Computação 13565-905 São Carlos Brasil

Abstract

This paper shows that genetic algorithms can be used as an optimization tool in conjunction with an instance-based learning method, to produce a combination which improves the performance the learning method could achieve on its own. Two instance-based methods are investigated in collaboration with genetic algorithms, namely k-NN and IB2.

We conducted a few experiments using a genetic algorithm for finding a ‘good’ weight vector for either learning algorithms. Classification results on three knowledge domains obtained using k-NN and IB2 modified by a weight vector found by a genetic algorithm, exceeds the performance of the instance-based methods on their own.

Keywords:Instance-Based Methods, Lazy Learning, Genetic Instance-Based Collaboration, Weighted NN, Weighted IB2

BibTex

@INPROCEEDINGS{de-nardin04:4,
                  AUTHOR       = {Luciana De Nardin and Maria do Carmo Nicoletti},
                  TITLE        = {A Genetic Instance-Based Collaborative Approach for Attribute Weightings},
                  BOOKTITLE    = {30ma Conferencia Latinoamericana de Informática (CLEI2004)},
                  YEAR         = {2004},
                  editor       = {Mauricio Solar and David Fernández-Baca and Ernesto Cuadros-Vargas},
                  pages        = {33--41},
                  address      = {},
                  month        = Sep,
                  organization = {Sociedad Peruana de Computación},
                  note         = {ISBN 9972-9876-2-0},
                  file         = {http://clei2004.spc.org.pe/es/html/pdfs/4.pdf}
}

PDF de este artículo
PDF de CLEI2004 (incluye todos los artículos)
Página principal CLEI 2004
Generado por Sociedad Peruana de Computación