基于AdaBoost的限制性貝葉斯組合分類器研究.pdf

1、北京交通大學碩士學位論文基于AdaBoost的限制性貝葉斯組合分類器研究姓名：李廣群申請學位級別：碩士專業(yè)：計算機應用技術指導教師：王志海20071201ABSlRACTInthedomainofDataMimng，oneofveryimportanttechnologiesisclassification11帕technologyofclassifiercombinationistocombineanumberofdifferents

2、ingleclassifierstogethertobecomeacompoundclassifierSO勰toimprovetheperformanceoforiginalclassifierOneofthetypicalalgorithmsbasedonBoostingisAdaBoosLRestrictedBayesianclassifieris0110ofthemosthotspotinclassificationtechnol

3、ogyresearchareaTIIispaperfirstlyintroducesconceptsandrelevantteehnblogiesofclassificationSecondlydescribesrelevanttechnologiesofclassifiercombinationincludingBaggin呂BoostingandSmc虹ngespeciallyanalyzesAdaBoostalgorithm’st

4、heoreticalbasisandkeystepsofBoostingtechnologyAfterelaboratingrelevantconceptsandtheoreticalbasisofrestrictedbayesianclassifieranalyzesthetheoreticalbasisandclassifiersstructureofNaivebayesClassifierTANClassifierHillClim

5、bingClassifierandSuperParentClassifiercomparingtheadvantagesanddisadvantagesofeachother鹋well，implementstheseclassifiersontheplatformofWekaOntheotherhandanalyzeshiddennaivebayesclassifierincludingconcepts，thestructureofcl

6、assifierandtheconstructionprocessofhiddennodes，describesprocessofthisalgorithmOnthebasisoftheresearchworkaboveweputforwardanewclassifiercombinationalgorithmcalledBoostTHNBItintroducesthe峨strcturetoHiddenNa／veBayesClassif

7、ieroneachleafnodeofwhichaHNBclassifierisconstructed；thesplittingapproachoftreenodeisaccordingtotrainingeITOFratesofHNBclassifierandifthetrainingerrol“rateissmalleraftersplitthansplitthisnodeotherwi∞donotsplitthisnodeWeca

8、lledthisclassifierTHNBaftermodificationWeusethealgorithmofAdaBoosttocombineclassifierswitllTHNBbeingthebasicclassifierAtlastcompar鶴newcombinedclassifierwithoriginalHNBclassifierandotherclassicalclassificationalgorithms，i

9、ncludingNaiveBayesclassifier刪classifieranddecisiontreealgorithmResultsandanalysisoftheexperimentshowthatthecombinationclassifiercanimproveclassificationpcrforlnanceoforiginalclassifieronmostdatasetsKEYWORDS：DataMing；Clas