资源描述:
《基于样本之间紧密度的模糊支持向量机方法》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、ISSN1000-9825,CODENRUXUEWE-mail:jos@iscas.ac.cnJournalofSoftware,Vol.17,No.5,May2006,pp.951−958http://www.jos.org.cnDOI:10.1360/jos170951Tel/Fax:+86-10-62562563©2006byJournalofSoftware.Allrightsreserved.∗基于样本之间紧密度的模糊支持向量机方法1,2+31张翔,肖小玲,徐光祐1(清华大学计算机科学与技术系,北京100084)2(长江大学
2、地球物理与石油资源学院,湖北荆州434023)3(武汉理工大学计算机科学与技术学院,湖北武汉430063)FuzzySupportVectorMachineBasedonAffinityAmongSamples1,2+31ZHANGXiang,XIAOXiao-Ling,XUGuang-You1(DepartmentofComputerScienceandTechnology,TsinghuaUniversity,Beijing100084,China)2(SchoolofGeophysicsandOilResources,Yangt
3、zeUniversity,Jingzhou434023,China)3(SchoolofComputerScienceandTechnology,WuhanUniversityofTechnology,Wuhan430063,China)+Correspondingauthor:Phn:+86-10-62782406,E-mail:xiang-zhang@tsinghua.edu.cn,http://www.tsinghua.edu.cnZhangX,XiaoXL,XuGY.Fuzzysupportvectormachinebased
4、onaffinityamongsamples.JournalofSoftware,2006,17(5):951−958.http://www.jos.org.cn/1000-9825/17/951.htmAbstract:SinceSVMisverysensitivetooutliersandnoisesinthetrainingset,afuzzysupportvectormachinealgorithmbasedonaffinityamongsamplesisproposedinthispaper.Thefuzzymembersh
5、ipisdefinedbynotonlytherelationbetweenasampleanditsclustercenter,butalsothoseamongsamples,whichisdescribedbytheaffinityamongsamples.Amethoddefiningtheaffinityamongsamplesisconsideredusingaspherewithminimumvolumewhilecontainingthemaximumofthesamples.Then,thefuzzymembersh
6、ipisdefinedaccordingtothepositionofsamplesinspherespace.Comparedwiththefuzzysupportvectormachinealgorithmbasedontherelationbetweenasampleanditsclustercenter,thismethodeffectivelydistinguishesbetweenthevalidsamplesandtheoutliersornoises.Experimentalresultsshowthatthefuzz
7、ysupportvectormachinebasedontheaffinityamongsamplesismorerobustthanthetraditionalsupportvectormachine,andthefuzzysupportvectormachinesbasedonthedistanceofasampleanditsclustercenter.Keywords:fuzzysupportvectormachine;affinity;classification摘要:针对传统支持向量机方法中存在对噪声或野值敏感的问题,提出
8、了一种基于紧密度的模糊支持向量机方法.在确定样本的隶属度时,不仅考虑了样本与类中心之间的关系,还考虑了类中各个样本之间的关系.通过样本之间的紧密度来描述类中各个样本之间的关系,利用包围同一类中样本的最小球半径大小来度量样