关键词: 支持向量机, 混合核核函数

Abstract: Mixed kernel function is not the right solution to the problem for selection of kernel function. It just transforms the problem to the selection of weight parameters. And we also need to determine the parameters of the two kernel functions respectively at the same time, which greatly increases the complexity of the algorithm and limits the generalization ability of support vec tor machine. In fact, the classification result depends on the parameters of kernel functions rather than itself. So the mixed kernel function actually dwells on the trivial. The empirical analysis shows that the proportion of common support vector, corresponding to different kernel functions, is very high. That means the linear combination of the kernel is not significant. It is an important reason that mixed kernel is unable to effectively improve the performance of classification.

Key words: Support Vector Machine, Mixed Kernel Function