Increasing Efficiency of Support Vector Machine using the Novel Kernel Function: Combination of Polynomial and Radial Basis Function

Support Vector Machine (SVM) is one of the most robust and accurate method amongst all the supervised machine learning techniques. Still, the performance of SVM is greatly influenced by the selection of kernel function. This research analyses the characteristics of the two well known existing kernel functions, local Gaussian Radial Basis Function and global Polynomial kernel function. Based on the analysis a new kernel function has been proposed which we call as “Radial Basis Polynomial Kernel (RBPK)”. The RBPK improves the learning as well as generalization capability of SVM. The performance of the proposed kernel function is illustrated on several datasets in comparison with single existing kernels. The result on different datasets from various domains has shown better learning and prediction ability of Support Vector Machine for the RBPK. Index Terms Support vector machine, kernel function, sequential minimal optimization, feature space, polynomial kernel, Radial Basis function