资源描述：
《统计模式识别中的维数削减与低损降维》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、统计模式识别中的维数削减与低损降维1),2)3)2)4)宋枫溪高秀梅刘树海杨静宇1)(哈尔滨工业大学深圳研究生院深圳518000)2)(炮兵学院二系合肥230031)3)(淮阴师范学院计算机系淮阴223001)4)(南京理工大学计算机系南京210094)摘要较为全面地回顾了统计模式识别中常用的一些特征选择、特征提取等主流特征降维方法,介绍了它们各自的特点及其适用范围,在此基础上,提出了一种新的基于最优分类器———贝叶斯分类器的可用于自动文本分类及其它大样本模式分类的特征选择方法———低损降维.在标准数据集
2、Reuters221578上进行的仿真实验结果表明,与互信息、χ2统计量以及文档频率这三种主流文本特征选择方法相比,低损降维的降维效果与互信息、χ2统计量相当,而优于文档频率.关键词维数削减;特征选择;特征抽取;低损降维;文本分类中图法分类号TP18DimensionalityReductioninStatisticalPatternRecognitionandLowLossDimensionalityReductionSONGFeng2Xi1),2)GAOXiu2Mei3)LIUShu2Hai2)YAN
3、GJing2Yu4)1)(ShenzhenGraduateSchool,HarbinInstituteofTechnology,Shenzhen518000)2)(NewStarResearchInstituteofAppliedTechnologyinHefeiCity,Hefei230031)3)(DepartmentofComputer,HuaiyinTeachersCollege,Huaiyin223001)4)(DepartmentofComputer,NanjingUniversityofSc
4、ienceandTechnology,Nanjing210094)AbstractFirst,authorsreviewtheprevailingfeatureselectionmethodssuchasExhaustiveSearch,GeneticAlgorithm,SequentialForwardFloatingSelection,andBestIndividualFeatures,andfeatureextractionapproachessuchasPrincipalComponentAnal
5、ysis,FisherDiscriminantA2nalysis,andProjectionPursuitforfeaturespacedimensionalityreductioninstatisticalpatternrecognition.Second,authorsdiscussthecharacteristicsandtheapplicabledomainsofallthesetechniques.Third,authorsproposeanovelfeatureselectionmethodb
6、asedonso2calledoptimalclassifier,Bayesianclassifier.Thenewfeatureselectionmethod,i.e.thelowlossdimensionalityreduction(LLDR),isappliedinautomatictextcategorizationandcomparedwiththeprevailingfeatureselectionmethodssuchasMutualInformation(MI),Chi2squareSta
7、tistic(CHI),andDocumentFrequency(DF)inautomatictextcategorization.ExperimentalresultsperformedonthewellknowndatasetReuters221578showthattheabilityfordimensionalityreductionofLLDRcomparedwiththoseofMIandCHI,andhigherthanthatofDF.ConsideringthatLLDRismoreco
8、mputationalefficientthanMIandCHI,LLDRisapromisingfeatureselectionmethodforauto2matictextcategorization.dimensionalityreduction;featureselection;featureextraction;lowlossdimensional2Keywordsityreduction;textcategoriz
