基于相异度核空间的支持向量机算法  

A dissimilarity based kernel space for support vector machines

作　　者：王萌[1,2] 孙树栋[1,2] 

机构地区：[1]西北工业大学系统集成与工程管理研究所,西安710072 [2]西北工业大学现代设计与集成制造技术教育部重点实验室,西安710072

出　　处：《系统工程理论与实践》2013年第6期1596-1600,共5页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(51075337)

摘　　要：提出基于相异度核空间的混合流形学习与支持向量机(SVMs)算法用于解决高维数据分类问题.该算法使用SVMs对流形学习降维后的低维嵌入做分类.通过加常数方法构造了衡量数据点相异度的特征空间.证明了该特征空间不仅能够衡量数据点之间的相异度,还满足Mercer条件对于支持向量机分类核空间的半正定性的要求.在仿真实验中,采用UCI机器学习数据库上标准分类数据集为样本.对比构造的相异度核空间与常用核函数如线性核、多项式核、高斯核在支持向量机分类时的分类精度来验证有效性和优越性.A dissimilarity based kernel space for hybrid manifold learning and support vector machines （SVMs） classification algorithm is proposed to solve the high-dimensional data classification problem. The SVMs classify the low-dimensional embedding of the manifold learning method in the algorithm. The dissimilarity based kernel space is constructed by using a constant additive method for SVMs. The constructed kernel space is guaranteed to be positive semi-definite in dissimilarity description and meet the Mercer condition. The simulation compares the kernel space with several common kernel including linear, polynomial and Gaussian kernel on data sets selected from UCI benchmark repository to validate its effectiveness and the dissimilarity based kernel space is an optimized kernel sp^ce/or hybrid manifold learning and SVMs classification problem.

关 键 词：等距映射 支持向量机 加常数 相异度 

分 类 号：TP391.5[自动化与计算机技术—计算机应用技术]