Training a Support Vector Classifier using a Cauchy-Laplace Product Kernel

The importance of the support vector machine and its applicability to a wide range of problems is well known. The strength of the support vector machine lies in its kernel. In our recent paper, we have shown how the Laplacian kernel overcomes some of the drawbacks of the Gaussian kernel. However this was not a total remedy for the shortcomings of the Gaussian kernel. In this paper, we design a Cauchy-Laplace product kernel to further improve the performance of the Laplacian kernel. The new kernel alleviates the deficiencies more effectively. During the experimentation with three data sets, it is found that the product kernel not only enhances the performance of the support vector machine in terms of classification accuracy but it results in obtaining higher classification accuracy for smaller values of the kernel parameter ?. Therefore the support vector machine gives smoother decision boundary and the results obtained by the product kernel are more reliable as it overcomes the problems of over fitting.