The Study On Fourier Kernel Function In Support Vector Machine

Support Vector Machine (SVM) is the main content of Statistics Learning Theory developed from 1990s. The kernel function is the crucial ingredient of SVM.Fourier kernel function is used in support vector machines because of its good properties sometimes, Many applications demonstrate that the performance of SVM with Fourier kernel is influenced greatly by the scal parameter q.In this paper, the author show how the parameter q affects the function of SVM with Fourier kernel. It is also proved that when qâ†'1, all the training samples are classified correctly by SVM with Fourier kernel, but the SVM can only get a constant function when qâ†'0; In both cases, SVM with Fourier kernel has little generalization ability. Experimental results validate the conclusion.Last, in this paper, six standadrd datas are used. Classification properties of SVM with Fourier kernel is examined by them, to the each standard data classified by SVM with Fourier kernel, the best parameter and the least error percent are found. they also are used to classify by SVM with Gauss kernel, However, not only calculate the error's percent, but also both of results are compared.