On domain knowledge and feature selection using a support vector machine

The basic principles of a support vector machine (SVM) are analyzed. The problem of feature selection while using an SVM is specifically addressed. An approach to constructing a kernel function which takes into account some domain knowledge about a problem and thus essentially diminishes the number of noisy parameters in high dimensional feature space is suggested. Its application to Texture Recognition is described. (C) 1999 Elsevier Science B.V. All rights reserved.