A kernel based feature selection method used in the diagnosis of wisconsin breast cancer dataset

Abstract

In this paper, a novel feature selection method called kernel F-score is applied for Breast cancer diagnosis. In this method, feature selection for removing the irrelevant/redundant features is achieved in high dimensional spaces than the original spaces. Basically, the datasets in the input space are moved to high dimensional kernel spaces for clear separation of nonlinearity through kernel functions. Then the F-score values for all the features in the kernel space are computed and mean kernel F-score value is set as the threshold for selection or rejection of features. The features lesser than the threshold are removed from feature space. The features above and equal to the threshold are selected for classification and used in the classification of benign and malignant cases using Support Vector Machines (SVM). The results obtained from Wisconsin Breast Cancer Dataset (WBCD) have been satisfied as it produced efficient results than F-score. So, we conclude kernel F-score with SVM for WBCD is promising than F-score with SVM. © 2011 Springer-Verlag.