剪枝和網(wǎng)格采樣相結(jié)合的非平衡數(shù)據(jù)集分類方法.pdf

1、獨創(chuàng)性聲明本人聲明所呈交的學位論文是本人在導(dǎo)師指導(dǎo)下進行的研究工作及取得的研究成果據(jù)我所知，除了文中特別加以標注和致謝的地方外，論文中不包含其他人已經(jīng)發(fā)表或撰寫過的研究成果，也不包含為獲得蝴其他教育機構(gòu)的學位或證書而使用過的材料。與我一同工作的同志對本研究所做的任何貢獻均已在論文中作了明確的說明并表示謝意學位論文作者簽名：掀簽字日期：幽p年、廣月，。日學位論文版權(quán)使用授權(quán)書本學位論文作者完全了解燃有關(guān)保留、使用學位論文的規(guī)定，有權(quán)保留并

2、向國家有關(guān)部門或機構(gòu)送交論文的復(fù)印件和磁盤，允許論文被查閱和借閱。本人授杈竊麴啦以將學位論文的全部或部分內(nèi)容編入有關(guān)數(shù)據(jù)庫進行檢搴，可以采用影印，縮印或掃描等復(fù)制手段保存、匯編學位論文，。(保密的學位論文在解密后適用本授權(quán)書)學位論文作者簽名：烈導(dǎo)師簽名：秀’老—弦3簽字日期妒p年』月，礦日簽字日期：o趴年鋤／曰學位論文作者畢業(yè)去向：工作單位：電話：通訊地址：郵編：AbstractImbalanceddatasetsclassifica

3、tionproblemiscommonproblemsinthefieldofpaRernrecognition，machinelearninganddataminingaswellasahotissueImbalanceddatasetisadatasetofcategoriesbecauseofthepresenceofskewnamelyakindofcategorysamplesmorethanothercategoriesof

4、sampleThetraditionalclassifiersinordertopursueahighrateofaccuracyfocusonclassificationaccuracyofthemajorityclasssamplesofImbalanceddatasets，ontheotherhandtheminorityclasssamplesofimbalanceddatasetsshouldbcconsideredbecau

5、seofthecostofclassificationanditstrueinformationTherefore，researchofImbalanceddataprocessingproblemisveryimportantAtpresentdomesticandforeignscholarshaveobtainedsomeachievementsindatapreprocessingandalgorithmsoftwolevela

6、boutimbalanceddatasetsclassificationproblemScholarsaretryingtoimprovethetraditionalalgorithmsandimprovetheclassificationperformanceinImbalanceddatasetonthealgorithmlevelInthedatapretreatmentlevel，scholarsgenerallyremovet

7、henegativesamplesofnoisedataandseparatefromtheclassificationofsurfacedatainundersampling，otherwisetheyaddnoisedatatooversamplingdatainordertobalanceInaword，manymethodsaredifferentondatareductionordataadditionInthispaper’

8、newsamplingmethodsaboutimbalanceddatasetsclassificationareconsideredinordertopreventthefromgeneralundersamplingmethodonthebasisofpreviousstudiesGridsamplingmethodbypruningputsforward，namelythemajorityclassofthesampleswil

9、lbedividedintoabsolutesafetydatadataofedgeandnoisedatabefore鰣dsamplingbasingonadaptiveboostingmethodtocarryOUtsamplingdatalearningTheartificialdataandtypicalUCIdatasetsareverifiedwithROCcurveastheevaluationcriterionbytes

10、tInlightoftestconclusion，theAUCvalueisgreaterthantheothertypesofalgorithms，whichshowsthemodelhasgoodperformanceTheothernewmethod，namelythemixedsamplingontheRandomSMOTEmethod，putforwardandisvalidbytestKeywords：Imbalancedd