多分類器系統(tǒng)中的組合方法及差異性度量研究.pdf

1、浙江大學(xué)碩士學(xué)位論文多分類器系統(tǒng)中的組合方法及差異性度量研究姓名：劉文瑤申請學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用指導(dǎo)教師：吳朝暉潘綱20050301浙江大學(xué)碩士學(xué)位論文多分類器系統(tǒng)中的組合方法及差異性度量研究～————————_h————_———_———^———————————————‘————————————————_—————————_——————————_’———●————_———_—_———～AbstractSincemultiple

2、classifiersystemscantosomeextentimprovetheperformanceofclassification，thetechniquehasbeenwidelyusedinvariousfieldsofpatternrecognitionResearchershavepropsosedanumberofcombiningmethods，SOit’SnecessarytomakeasurveyonthemIn

3、addition，therecentresearchor／multipleclassifiersystemsisriolongerlimitedtoproposingorenhancingthecombiningmethodsMoreandmoreattentionispaidfortherelationshipamongbaseclassifiers，especiallythediversityinclassifierensemble

4、sResearcherswanttofindaproperdiversitymeasure，whoseabilityinpredictingtheperformanceofclassifercombiningcanbeusedtoassisttheconstructionofmultipleclassifiersystemsThispaperstudiesthesetwomainproblemsinmul邱leclassifiersys

5、tems：combiningmethodsanddiversitymeasuresThemaincontributionsoftheworkareasfollows：FirstlyasurveyOilthecombiningmethodsinmultipleclassifiersystemsisgiven，inwhichtheadvantagesanddisadvantagesofeachmethodaresummarizedInadd

6、ition，wedesignsomeexperimentstocomparethemethodsSecondlyanewkindofdiversitymeasllreforclassifierensemblesnamedEPDisproposedTheexperimentsOriUCIMachineLearningRepositoryprovethat，comparedtoexistingmeaslLl“es，EPDshowsstron

7、gerabilityinpredictingtheperformanceofmultipleclassifiersystemsThirdlya11algorithmofensemblethinningbasedonEPDisdevelopedandusedtoconstructanimprovedfacerecognitionsystemTheexperimentsOriAT&Tfacedatabaseprovethat，usil培th