A novel approach to select most effective attributes for SVM algorithm

Abstract

Support Vector Machine (SVM) is one of the most popular machine learning algorithms for pattern recognition of a specific dataset. The percentage of accuracy from a defined SVM model greatly depends on the selection of appropriate attributes for SVM model. But the most effective attributes selection for SVM algorithm is one of the most difficult tasks for any kind of data classification. A mathematical model is proposed in this paper through which effectiveness of attributes for SVM model can be calculated. The validity of this SVM model is justified by comparing the effectiveness of a SVM model with the rate of pattern recognition for corresponding SVM model. Linear, Radial Basis Function (RBF), Polynomial Kernel SVM algorithms are used for pattern recognition. The percentage of accuracy of pattern recognition increases with the effectiveness of SVM model. The range of value of effectiveness of SVM model is 0 to <x. We have tested our proposed algorithm for noninvasive Brain Computer Interface (BCI) system. The proposed mathematical model is applicable for any other linear or nonlinear system.