ensemble methods in data mining (2010)

《ensemble methods in data mining (2010)》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、EnsembleMethodsinDataMining:ImprovingAccuracyThroughCombiningPredictionsSynthesisLecturesonDataMiningandKnowledgeDiscoveryEditorRobertGrossman,UniversityofIllinois,ChicagoEnsembleMethodsinDataMining:ImprovingAccuracyThroughCombiningPredictionsGiovanniSeniandJohnF.E

2、lder2010ModelingandDataMininginBlogosphereNitinAgarwalandHuanLiu2009Copyright©2010byMorgan&ClaypoolAllrightsreserved.Nopartofthispublicationmaybereproduced,storedinaretrievalsystem,ortransmittedinanyformorbyanymeanselectronic,mechanical,photocopy,recording,oranyoth

3、erexceptforbriefquotationsinprintedreviews,withoutthepriorpermissionofthepublisher.EnsembleMethodsinDataMining:ImprovingAccuracyThroughCombiningPredictionsGiovanniSeniandJohnF.Elderwww.morganclaypool.comISBN:9781608452842paperbackISBN:9781608452859ebookDOI10.2200/S

4、00240ED1V01Y200912DMK002APublicationintheMorgan&ClaypoolPublishersseriesSYNTHESISLECTURESONDATAMININGANDKNOWLEDGEDISCOVERYLecture#2SeriesEditor:RobertGrossman,UniversityofIllinois,ChicagoSeriesISSNSynthesisLecturesonDataMiningandKnowledgeDiscoveryPrint2151-0067Elec

5、tronic2151-0075EnsembleMethodsinDataMining:ImprovingAccuracyThroughCombiningPredictionsGiovanniSeniElderResearch,Inc.andSantaClaraUniversityJohnF.ElderElderResearch,Inc.andUniversityofVirginiaSYNTHESISLECTURESONDATAMININGANDKNOWLEDGEDISCOVERY#2MMorgan&cLaypoolpubli

6、shers&CABSTRACTEnsemblemethodshavebeencalledthemostinfluentialdevelopmentinDataMiningandMachineLearninginthepastdecade.Theycombinemultiplemodelsintooneusuallymoreaccuratethanthebestofitscomponents.Ensemblescanprovideacriticalboosttoindustrialchallengesfrominvestment

7、timingtodrugdiscovery,andfrauddetectiontorecommendationsystemswherepredictiveaccuracyismorevitalthanmodelinterpretability.Ensemblesareusefulwithallmodelingalgorithms,butthisbookfocusesondecisiontreestoexplainthemmostclearly.Afterdescribingtreesandtheirstrengthsandw

8、eaknesses,theauthorsprovideanoverviewofregularizationtodayunderstoodtobeakeyreasonforthesuperiorper-formanceofmodernensemblingalgorithms.Thebookc