资源描述:
《离群模糊核聚类算法》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、1000-9825/2004/15(07)1021©2004JournalofSoftware软件学报Vol.15,No.7∗离群模糊核聚类算法1,21+2,3沈红斌,王士同,吴小俊1(江南大学信息学院,江苏无锡214036)2(华东船舶工业学院计算机系,江苏镇江212003)3(中国科学院沈阳自动化研究所机器人学重点实验室,辽宁沈阳110015)FuzzyKernelClusteringwithOutliers1,21+2,3SHENHong-Bin,WANGShi-Tong,WUXiao-Jun1(SchoolofInformation,SouthernYangtseUniversi
2、ty,Wuxi214036,China)2(DepartmentofComputer,EastChinaShipbuildingInstitute,Zhenjiang212003,China)3(RoboticsLaboratory,ShenyangInstituteofAutomation,TheChineseAcademyofSciences,Shenyang110015,China)+Correspondingauthor:E-mail:wxwangst@yahoo.com.cn,http://www.pami.sjtu.edu.cnReceived2003-08-11;Accep
3、ted2003-10-08ShenHB,WangST,WuXJ.Fuzzykernelclusteringwithoutliers.JournalofSoftware,2004,15(7):1021~1029.http://www.jos.org.cn/1000-9825/15/1021.htmAbstract:Outliersaredatavaluesthatlieawayfromthegeneralclustersofotherdatavalues.Itmaybethatanoutlierimpliesthemostimportantfeatureofadataset.Inthisp
4、aper,anewfuzzykernelclusteringalgorithmispresentedtolocatethecriticalareasthatareoftenrepresentedbyonlyafewoutliers.Throughmercerkernelfunctions,thedataintheoriginalspacearefirstlymappedtoahigh-dimensionalfeaturespace.Thenamodifiedobjectivefunctionforfuzzyclusteringisintroducedinthefeaturespace.A
5、nadditionalweightingfactorisassignedtoeachvectorinthefeaturespace,andtheweightvalueisupdatedusingtheiterativefunctionsderivedfromtheobjectivefunction.Thefinalweightofadatumrepresentsakindofrepresentativenessofthecorrespondingdatum.Withtheseweights,theexpertscanidentifytheoutlierseasily.Thesimulat
6、ionsdemonstratethefeasibilityofthismethod.Keywords:outlier;fuzzy;kernelfunction;featurespace;clusteringalgorithm摘要:一般说来,离群点是远离其他数据点的数据,但很可能包含着极其重要的信息.提出了一种新的离群模糊核聚类算法来发现样本集中的离群点.通过Mercer核把原来的数据空间映射到特征空间,并为特征空间的每∗SupportedbytheJiangsuKeyLaboratoryofComputerInformationTechnology(江苏省计算机信息技术重点实验室开放课题
7、);theNationalKeyLaboratoryforNovelSoftwareTechnologyofNanjingUniversity(南京大学计算机软件新技术国家重点实验室开放课题);theJiangsuNaturalScienceFoundationofChinaunderGrantNo.BK2003017(江苏省自然科学基金)作者简介:沈红斌(1979-),男,江苏句容人,博士,主要研究领域为模糊人工智能,数据挖掘;王