Abstract
In this paper, a universal kernel function is introduced that could improve the classification accuracy of Support Vector Machines (SVMs) for both linear and nonlinear data sets. A class of universal kernel functions on the basis of the properties of the common kernels is proposed, which can find numerous applications in practice. The proposed kernels satisfy Mercer's condition and can be used for generating the most established kernels such as Gaussian Radial Function (GRF), Polynomial Radial Basis Function (PRBF), and Polynomial Exponential Radial Function (PERF) of SVMs. The SVM with the universal kernel is experimentally applied to a variety of nonseparable data sets with several attributes, leading to good classification accuracy in nearly all the data sets, especially those of high dimensions. The use of the universal kernel results in a better performance than those with established kernels.