Kernel-target Alignment Based Multiple Kernel One-class Support Vector Machine

Abstract

One-class support vector machine is a hot research topic in the domain of machine learning. It is currently widely used to deal with one-class classification problems or classification problems of class imbalance data, and has good performance in many practical applications. A key problem of one-class support vector machine is the selection of its kernel function and parameters, which has a vital impact on the final performance of the classifier. At present, there is no unified method for how to select the appropriate kernel function and its parameters. In order to solve this problem, the multiple kernel method is introduced into the one-class support vector machine. i.e., a combined kernel is used to replace a single kernel in the one-class support vector machine, where the combined kernel is obtained by weighted summation of several basic kernels, and the kernel weight is calculated by the kernel-target alignment. The experimental results on UCI database show that this method can effectively save training time and solve the selection of kernel parameter and its parameters based on high classification performance.