An Information Criterion for Variable Selection in Support Vector Machines

33 Pages Posted: 19 Feb 2008

See all articles by Gerda Claeskens

Gerda Claeskens

KU Leuven - Department of Economics

Christophe Croux

KU Leuven - Faculty of Business and Economics (FEB)

Johan Van Kerckhoven

Katholieke Universiteit Leuven (KUL)

Date Written: 2007

Abstract

Using support vector machines for classification problems has the advantage that the curse of dimensionality is circumvented. However, it has been shown that even here a reduction of the dimension of the input space leads to better results. For this purpose, we propose two information criteria which can be computed directly from the definition of the support vector machine. We assess the predictive performance of the models selected by our new criteria and compare them to a few existing variable selection techniques in a simulation study. Results of this simulation study show that the new criteria are very competitive compared to the others in terms of out-of-sample error rate while being much easier to compute. When we repeat this comparison on a few real-world benchmark datasets, we arrive at the same findings.

Keywords: Classification, Criteria, Error rate, Information, Information criterion, IT, Model, Models, Performance, Problems, Selection, Simulation, Space, Studies, Supervised classification, Support vector machine, Variable selection

Suggested Citation

Claeskens, Gerda and Croux, Christophe and Van Kerckhoven, Johan, An Information Criterion for Variable Selection in Support Vector Machines (2007). Available at SSRN: https://ssrn.com/abstract=1094652 or http://dx.doi.org/10.2139/ssrn.1094652

Gerda Claeskens (Contact Author)

KU Leuven - Department of Economics ( email )

Leuven, B-3000
Belgium

Christophe Croux

KU Leuven - Faculty of Business and Economics (FEB) ( email )

Naamsestraat 69
Leuven, B-3000
Belgium

Johan Van Kerckhoven

Katholieke Universiteit Leuven (KUL) ( email )

Oude Markt 13
Leuven, Vlaams-Brabant 3000
Belgium

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
120
Abstract Views
838
Rank
419,528
PlumX Metrics