A Novel Method of Feature Selection based on SVM

A novel method of feature selection combined with sample selection is proposed to select discriminant features in this paper. Based on support vector machine trained on training set, the samples excluding the misclassified samples and support vector samples are used to select informative features during the procedure of recursive feature selection. The feature selection method is applied to seven datasets, and the classification results of the selected discriminant features show that the method is effective and reliable for selecting features with high classification information.