Neural Network Nonlinear Regression Modeling and Information Criteria
by
Tomohiro Ando
Graduate School of Mathematics, Kyushu University, Fukuoka, Japan
Coauthors: Sadanori Konishi (Graduate School of Mathematics, Kyushu University, Fukuoka, Japan)

Neural networks have received considerable attention as useful tools for analyzing data with complex structure. We consider the problem of constructing nonlinear regression models, using multilayer perceptrons and radial basis function networks with the help of the technique of regularization. Crucial issues in the model building process are the choices of the number of basis functions, the number of hidden units and a regularization parameter. Ando, Konishi and Imoto (2001) proposed a nonlinear regression modeling strategy based on radial basis function networks, giving an information-theoretic criterion for the evaluation of a constructed model under model misspecification. We investigate the properties of nonlinear regression modeling strategies based on the multilayer perceptron and radial basis function networks. The simulation results show that radial basis function network model performs well, and in practice it has clear advantages in comparison with the multilayer perceptron network model.

Date received: September 17, 2001

Copyright © 2001 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # caid-19.