A method for improving the generalisation performance of constructive radial basis function (RBF) networks is proposed. Experiments using three image datasets are presented. The results show that the proposed method considerably improves performance of constructive RBF networks, outperforms multilayer perceptrons and AdaBoost and achieves comparable performance to support vector machines in these datasets.
References
-
-
1)
-
M. Berthold ,
J. Diamond
.
Constructive training of probabilistic neural networks.
Neurocomputing
,
167 -
183
-
2)
-
Ou, Y.-Y., Chen, C.-Y., Hwang, S.-C., Oyang, Y.-J.: `Expediting model selection for support vector machines based on data reduction', Proc. IEEE Conf. on Systems, Man and Cybernetics, 2003, 1, p. 786–791.
-
3)
-
Kaynak, C., Alpaydin, E.: `Multistage cascading of multiple classifiers: one man's noise in another man's data', Proc. 17th Int. Conf. on Machine Learning, 2000, p. 455–462.
-
4)
-
Blake, C., Merz, C.: `UCI repository of machine learning databases', 1998, http://www.ics.uci.edu/~mlearn/MLRepository.
-
5)
-
E.L. Allwein ,
R.E. Schapire ,
Y. Singer
.
Reducing multiclass to binary: a unified approach for margin classifiers.
J. Mach. Learn. Res.
,
113 -
141
-
6)
-
J. Paetz
.
Reducing the number of neurons in radial basis function networks with dynamic decay adjustment.
Neurocomputing
-
7)
-
D. Michie ,
D.J. Spielgelhalter ,
C.C. Taylor
.
(1994)
Machine learning, neural and statistical classification.
-
8)
-
M. Berthold ,
J. Diamond ,
G. Tesauro
.
(1995)
Boosting the performance of RBF networks with dynamic decay adjustment, Advances in neural information processing systems.
-
9)
-
R. Rifkin ,
A. Klautau
.
In defense of one-vs-all classification.
J. Mach. Learn. Res.
,
101 -
141
-
10)
-
Oliveira, A.L.I., Neto, F.B.L., Meira, S.R.L.: `Improving RBF-DDA performance on optical character recognition through parameter selection', Proc. Int. Conf. on Pattern Recognition, 2004, Cambridge, UK, 4, p. 625–628.
http://iet.metastore.ingenta.com/content/journals/10.1049/el_20057296
Related content
content/journals/10.1049/el_20057296
pub_keyword,iet_inspecKeyword,pub_concept
6
6