资源描述：
《03-基于最大熵原理的空间特征选择方法》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、1000-9825/2003/14(09)1544©2003JournalofSoftware软件学报Vol.14,No.9∗基于最大熵原理的空间特征选择方法1+1,211,2宋国杰,唐世渭,杨冬青,王腾蛟1(北京大学计算机科学技术系,北京100871)2(北京大学视觉与听觉信息处理国家重点实验室,北京100871)ASpatialFeatureSelectionMethodBasedonMaximumEntropyTheory1+1,211,2SONGGuo-Jie,TANGShi-Wei,YANGDong-Qing,WANGTeng-Jiao1(Departmentof
2、ComputerScienceandTechnology,PekingUniversity,Beijing100871,China)2(NationalLaboratoryonMachinePerception,PekingUniversity,Beijing100871,China)+Correspondingauthor:Phn:86-10-62763510,E-mail:sgj@db.pku.edu.cnhttp://db.cs.pku.edu.cnReceived2002-08-09;Accepted2002-12-23SongGJ,TangSW,YangDQ,Wa
3、ngTJ.Aspatialfeatureselectionmethodbasedonmaximumentropytheory.JournalofSoftware,2003,14(9):1544~1550.http://www.jos.org.cn/1000-9825/14/1544.htmAbstract:Featureselectionhasanimportantapplicationinthefieldofpatternrecognitionanddataminingetc.However,inrealworlddomains,iftherearespatialdata
4、operatedintheapplication,theperformanceoffeatureselectionwillbedecreasedbecauseofwithoutconsideringthecharacteristicofspatialdata.Inthispaper,afeatureselectionmethodfromthepointofthecharacteristicofspatialdata,namedMEFS(maximumentropyfeatureselection),isproposed.Basedonthetheoryofmaximumen
5、tropy,MEFSusesmutualinformationandZ-testtechnologies,andtakestwo-stepmethodtoexecutefeatureselection.Thefirststepispredicateselection,andthesecondstepistochooserelevantdatasetcorrespondingtoeachpredicate.Atlast,theexperimentsbetweenfeatureselectionalgorithmsMEFSandRELIEF,andbetweenID3class
6、ificationalgorithmandclassificationalgorithmbasedonMEFSarecarriedout.TheexperimentalresultsshowthattheMEFSalgorithmnotonlysavesfeatureselectionandclassificationtime,butalsoimprovesthequalityofclassification.Keywords:spatialdatamining;spatialfeatureselection;maximumentropytheory;mutualinfor
7、mation;decisiontree摘要:特征选择在模式识别和数据挖掘等领域都有十分广泛的应用.然而,当涉及空间数据时,由于传统特征选择方法没有很好地考虑数据的空间特性,所以会导致特征选择结果性能下降.从空间数据本身的特性出发,提出一种特征选择方法MEFS(maximumentropyfeatureselection).MEFS在基于最大熵原理的基础上,运用互信息和Z-测试技术,采用两步方法进行空间特征选择.第1步,空间谓词选择;第2步,选择与每个空间谓词对应的相关∗SupportedbytheFoundatio
