[R] variable selection with support vector machines (SVM)

Serge Zaugg sezaugg at gmx.ch
Tue Nov 21 15:50:12 CET 2006


I am using support vector machine (from package kernlab) for a classification task (with RBF-Kernel). My data has dozens of variables and I need to identify which variables contribute most to the classification performance. 

What I did so far is comparing the classification performance (measured for example with the proportion of misclassified cases) of different sets of variables with cross-validation. Unfortunately this is very slow and doing, for example, a backward variable selection procedure will take half a day with my data.

This raises 3 interrelated questions:

Does someone know an alternative way to perform variable selection in the context of SVM-classification ?  

Does someone know of an R-function that automatizes variable selection for SVM ?

Is there a way to quantify the contribution of every single variable to the classification performance ?   (I guess there is no short answer to this but I would also be very happy on references of good articles or books on this topic)

Thanks a lot in advance,

Serge Zaugg

Serge Zaugg
Gotthelfstrasse 14
CH-3013 Bern

sezaugg at gmx.ch
mobile: +41 (0)77 401 97 93
phone:  +41 (0)31 331 06 03

"Ein Herz für Kinder" - Ihre Spende hilft! Aktion: www.deutschlandsegelt.de
Unser Dankeschön: Ihr Name auf dem Segel der 1. deutschen America's Cup-Yacht!

More information about the R-help mailing list