文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Robust probabilistic PCA with missing data and contribution analysis for outlier detection
侵权投诉

Robust probabilistic PCA with missing data and contribution analysis for outlier detection

Robust probabilistic PCA with missing data and contribution analysis for outlier detection

ComputationalStatisticsandDataAnalysis53(2009)

Robust probabilistic PCA with missing data and contribution analysis for outlier detection

3706–3716

ContentslistsavailableatScienceDirect

ComputationalStatisticsandDataAnalysis

journalhomepage:

Robust probabilistic PCA with missing data and contribution analysis for outlier detection

doc.guandang.net

RobustprobabilisticPCAwithmissingdataandcontributionanalysisforoutlierdetection

TaoChena, ,ElaineMartinb,GaryMontagueb

a

bSchoolofChemicalandBiomedicalEngineering,NanyangTechnologicalUniversity,Singapore637459,SingaporeSchoolofChemicalEngineeringandAdvancedMaterials,NewcastleUniversity,NewcastleuponTyne,NE17RU,UK

articleinfoabstract

Principalcomponentanalysis(PCA)isawidelyadoptedmultivariatedataanalysis

technique,withinterpretationbeingestablishedonthebasisofbothclassicallinear

projectionandaprobabilitymodel(i.e.probabilisticPCA(PPCA)).RecentlyrobustPPCA

models,byusingthemultivariatet-distribution,havebeenproposedtoconsiderthe

situationwheretheremaybeoutlierswithinthedataset.Thispaperpresentsan

overviewoftherobustPPCAtechnique,andfurtherdiscussestheissueofmissingdata.

Anexpectation-maximization(EM)algorithmispresentedforthemaximumlikelihood

estimationofthemodelparametersinthepresenceofmissingdata.Whenapplying

robustPPCAforoutlierdetection,acontributionanalysismethodisproposedtoidentify

whichvariablescontributethemosttotheoccurrenceofoutliers,providingvaluable

informationregardingthesourceofoutlyingdata.Theproposedtechniqueisdemonstrated

onnumericalexamples,andtheapplicationtooutlierdetectionanddiagnosisinan

industrialfermentationprocess.

©2009ElsevierB.V.Allrightsreserved.Articlehistory:Received9March2009Accepted20March2009Availableonline31March2009

1.Introduction

Principalcomponentanalysis(PCA)(Jolliffe,2002)isageneralmultivariatestatisticalprojectiontechniquefordimensionreduction,andithasseenawidespectrumofapplicationsinvariousareas,includingexploratorydataanalysis,patternrecognition,qualitymonitoringandcontrol.ThetraditionalapproachtotheimplementationofPCAisbasedonthelinearprojectionoftheoriginaldataontoaspacewherethevarianceismaximized.Let{xn,n=1,...,N}bethed-dimensionaldataset,thenthefirststepinPCAistocomputethesamplecovariancematrix,S,oforderd×d.TheeigenvectorswjandeigenvaluesλjofSarethencalculated(j=1,...,d).Byretainingthoseeigenvectorscorrespondingtotheqlargesteigenvalues,theq-dimensionalPCAscorevectors,tn,arecalculatedas:tn=WT(xn µ),whereW=(w1,...,wq),andµisthemeanofthedataset.Thereforetheoriginaldatacanberepresentedasalinearcombinationofthescoresplusanoisevector:xn=Wtn+µ+en.

MorerecentlyTippingandBishop(1999b)proposedaprobabilisticformulationofPCA(PPCA)fromtheperspectiveofaGaussianlatentvariablemodel.ThePPCAmodelisrealizedbyspecifyingaGaussiannoisemodele~G(0,σ2I),whichimpliesthattheconditionaldistributionofdatagivenPCAscoresis:x|t~G(Wt+µ,σ2I).ByadoptingapriorGaussiandistributionforthePCAscorevector,t~G(0,I),themarginaldistributionofthedataxisalsoshowntobeGaussian:x~G(µ,WWT+σ2I).PPCAdegeneratesintotraditionalPCAifσ2→0.WithinthePPCAframework,theprincipalcomponentsareessentiallythemaximumlikelihoodestimatesofthemodelparameters,whichcanbeimplementedusingCorrespondingauthor.Tel.:+6565138267;fax:+6567947553.

E-mailaddress:chentao@ntu.edu.sg(T.Chen).

0167-9473/$–seefrontmatter©2009ElsevierB.V.Allrightsreserved.

doi:10.1016/j.csda.2009.03.014

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email