Improving SVM Classification by Feature Weight Learning

This paper presents a new feature weighting method to improve the performance of support vector machine (SVM). The basic idea of this method is to translate the feature weight learning into the problem of choosing a kernel suitable for SVM classification. In more detail, this method tunes the width parameters of Gaussian ARD (Automatic Relevance Determination) kernel via optimizing a kernel evaluation criterion, i.e., kernel polarization. By using gradient ascent technique, each learned parameter indicates the relative importance of the corresponding feature. The proposed method is demonstrated with some UCI machine learning benchmark examples.