为弥补具有径向基函数(radial basis function,RBF)核的支持向量机(support vector machine,SVM)的不足,即如何获得合适的RBF核并且包含更多的数据信息,提出一个关于核聚类的有效方法.该方法通过关于核数目、中心、宽度的一些公式来生成核.这个生成方法可以容易、合理、优化地获得核.在美国邮政手写数字数据集(记为USPS)和字母识别数据集(记为Letter)上的试验可以验证,具有这些新生成的RBF核的SVM拥有更少更合理的核以及更高的分类性能.通过这个新提出的方法,可以处理更多的大规模分类问题.%In order to make up for the disadvantage of support vector machine (SVM) with radial basis function (RBF) kernels,i.e.,how to get feasible RBF kernels with more data information,an effective method about kernel clustering is proposed.This method generates kernels with some formulas about the kernel number,kernel center and kernel width.This generation method can get kernels easily,reasonably and optimally.Tests on the United States Postal Service handwritten digit data set (called USPS) and the letter recognition data set (called Letter) validate that SVM with these new generated RBF kernels has fewer and more reasonable kernels and higher classification accuracy.With this new proposed method,the more large-scale classification problems can be processed.
展开▼
