论文部分内容阅读
利用矩阵能极分解生成一个对称半正定矩阵的特性,本文对RBF核进行极分解,并结合全局多项式核,构造一个性能较好的混合核函数.然后对该混合核设定两个权值,使之达到较好的性能.在UCI数据库中的数据集上进行实验,采用基于极分解下的混合核,来与RBF核进行比较.结果表明,使用混合核的SVM,其支持向量的个数少、分类错误低,并有较好的训练速度.还进一步发现,在大多数据集上,该混合核有效抑制了局部核函数RBF所引起的预测输出波动.
In this paper, we decompose the RBF kernel and combine it with the global polynomial kernel to construct a mixed kernel function with better performance. Then we set two weights for the mixed kernel, So as to achieve better performance.Experiments on datasets in UCI database are carried out to compare with RBF kernels based on the hybrid kernels under extreme decomposition.The results show that the number of support vectors Less misclassification errors and better training speed.It is further found that in most data sets, the hybrid kernel effectively suppresses the prediction output fluctuation caused by RBF of local kernel function.