基于傅立叶核与径向基核的支持向量机性能之比较

A performance comparison of SVMs based on Fourier kernel and RBF kernel

支持向量机(SVMs)是由Vapnik提出的一种建立在统计学习理论上的新方法。这种方法被深入地研究并广泛应用在诸如分类和回归问题上。由于其基于结构风险最小化的机理,因此相对于其他的经典方法有着更好的泛化特性,其中核函数的选择对支持向量机的性能有着很大的影响。深入地研究了基于傅立叶核函数的支持向量机的特性,得出在某些特殊的情况下,基于傅立叶核函数的支持向量机的性能要好于基于RBF核的支持向量机。最后的仿真对其进行了比较验证。

Support Vector Machines （SVMs） a new way based on statistical theory, was proposed by Vapnik. It has been well studied and widely applied to the classification and regression. According to the Structural Risk Minimization （SRM） principle, SVM has a better performance on generalization contrast than classic methods. The choice of kernel function of SVM is key factor in the performance. In this paper, Fourier kernel is studied. A conclusion is drawn that, in some area, SVM based on Fourier kernel is better than that based on RBF kernel. At last, a simulation is conducted to illustrate it.