基于聚類的潛在缺陷模塊序列預(yù)測.pdf

1、東南大學(xué)碩士學(xué)位論文基于聚類的潛在缺陷模塊序列預(yù)測姓名：楊明申請學(xué)位級別：碩士專業(yè)：計算機科學(xué)與技術(shù)；計算機軟件與理論指導(dǎo)教師：周曉宇201106東南大學(xué)碩士學(xué)位論文AbstractTraditionalpredictionmodelsformodulefaultpronenessrankingusesupeⅣisedmachineleamingmethodsPreviousresearchshowsthatthesemethodsha

2、veagoodperfo肋anceHowever’suchmodelsarebasedonhistorcaJdefectinfo徹ationFornewlydevelopedsoftwarcsystems，weareunabJetoapplysuchmodelsdirectly；ashisloricaldefectinfb咖ationisunavailableInordertodealwiththisproblem，thispaperp

3、roposesapredictionmodelf6rmodulefaultpronenessrankingbasedonclusteringmethodsTheproceduretobuildthismodelconsistsofthefollowingthreestepsAtthe6rststep，weuseaclusteringmethodandstTllcturalmetricstoclusterthemodulesofthesy

4、stemAtthesecondstep，wefindthelargestcluster，andputitonthetopofthelistandconsideritscentroidpointasthebasepointThen，f0reachcluster，weconlputethedistanceofitscentroidpointtothebasepointandsortallclustersaccordingtotheabove

5、distanceThecloserthedistancetothebasepoint，thehigherthepositiontheclusterwillbeAtthethirdstep，wecomputethedistanceofinstancesinotherclustertotheirowncentroidpointandsontheseinstancesaccordin2lotheabovedistanceInordertoev

6、aIuatetheefflectivenessofthemodel，weconductexper主mentsusin2threeversionsofEclipseTheresultshowsthatourmethodisbetterthanthreetypesofbenchmarkmodelsThenbenchmarkmodelarerandommodel，predictionmodelsbasedonfoursuperVisedmac

7、hineleamin2method(BinaryLogisticregression，NaiveBayes，ArtificialNeuralNetworl【andJ48)andmanualupmodel(sortmodulesinanascendingorderaccordin2totheirSLOC)7rtlethesismakesthefollowingcontributions：(1)Weuseunsupervisedmachin

8、eleamingmethodstobuildafaultpronenessrankingmodelformodulesSuchamodelisbasedonsourcecodeanddoesnotneedhistoricaldefectinfbmation：(2)WeempiricallyValidatetheproposedfaultpronenessrankingmodelusingthreeversionsofalargescal