多尺度高斯核支持向量机算法

Multi-scale Gaussian Kernel Algorithm Based on Support Vector Machine

针对支持向量机中单尺度高斯核算法存在局部风险的问题,提出一种基于核排列的多尺度高斯核算法。利用核排列这一度量标准来选择高斯核函数的尺度,并把多个弱分类器聚集成一个强分类器得到多尺度高斯核,从而构造支持向量机模型。利用UCI数据集Iris Plants、 Wine Recognition等仿真实验结果表明:所提出的基于核排列的多尺度高斯核算法比传统的单尺度高斯核算法具有更高的分类准确率。

In support vector machine,for the phenomenon of local risk in the application of single-scale Gaussian kernel algorithm,a multi-scale Gaussian kernel algorithm based on kernel arrangement was proposed.The kernel target alignment was used to select the scale of Gaussian kernel function,and multi weak classifiers were integrated into a strong classifiers to get the multi-scale Gaussian kernel,thereby the support vector machine was constructed.The experimental results of UCI dataset Iris Plants,Wine Recognition etc.show that the proposed multi-scale Gaussian kernel algorithm based on kernel target alignment has higher classification accuracy than the traditional single-scale Gaussian kernel function.