Predictive Subset Selection using Regression Trees and RBF Neural Networks Hybridized with the Genetic Algorithm

---

AKBİLGİÇ O., Bozdogan H.

EUROPEAN JOURNAL OF PURE AND APPLIED MATHEMATICS, vol.4, no.4, pp.467-485, 2011 (Peer-Reviewed Journal)

• Publication Type: Article / Article
• Volume: 4 Issue: 4
• Publication Date: 2011
• Journal Name: EUROPEAN JOURNAL OF PURE AND APPLIED MATHEMATICS
• Journal Indexes: TR DİZİN (ULAKBİM)
• Page Numbers: pp.467-485
• Istanbul University Affiliated: Yes

Abstract

 

In this paper we develop a novel nonparametric predictive subset regression modeling procedure

that involves a combination of regression trees with radial basis function (RBF) neural networks

hybridized with the genetic algorithm (GA) to carry out the subset selection of the best predictors. We

use the information-theoretic measure of complexity (ICOMP) criterion of [5, 6, 7, 8] as our fitness

function to choose the best approximating radial basis functions and to choose the best subset of predictors

with the GA. To avoid the potential singularities in the design matrix, we combine our model

with analytical global ridge regression for regularization. On the other hand, estimation and prediction

performance of model also taken into account for best subset chosen.