A Hybrid Kernel Support Vector Machine with Feature Selection for the Diagnosis of Diseases

Abstract

Feature selection (FS) is very important to speed up learning, generalization ability and computational efficiency to improve concept quality. In this paper, hybrid kernel Support Vector Machine (SVM) with feature selection is proposed to increase the performance of classification. The modified hybrid kernel SVM feature selection approach is used to increase the classification accuracy and generalization ability of filter methods by adding a particular learning algorithm in the selection procedure and narrowing the searching space to increase the efficiency of wrapper approaches. Moreover, a data preprocessing technique that improves classification accuracy and generalization ability by reducing noisy data. To verify the feasibility and efficiency of this proposed method, we used two different nonlinear breast cancer and heart disease datasets with different characteristics. The simulation results illustrate that the hybrid kernel SVM with feature selection is promising and effective while comparing with existing methods.