用r实现随机森林的分类与回归

《用r实现随机森林的分类与回归》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、用R实现随机森林的分类与回归第五届中国R语言会议北京2012李欣海用R实现随机森林的分类与回归ApplicationsofRandomForestusingRClassificationandRegression李欣海中科院动物所邮件:lixh@//0>.主页:////.博客:////.微博:////.第五届中国R语言会议北京2012李欣海随机森林简介RandomForest////.an-introduction-to-data-mining-for-marketing-and-business-intelligence/Rando

2、mForestisanensembleclassifierthatconsistsofmanydecisiontreesItoutputstheclassthatisthemodeoftheclass'soutputbyindividualtreesBreiman2001Itdealswith“smallnlargep”-problems,high-orderinteractions,correlatedpredictorvariables.Breiman,L.2001.Randomforests.MachineLearning45:

3、5-32.Beingcited6500timesuntil20123/25第五届中国R语言会议北京2012李欣海随机森林简介History////.an-introduction-to-data-mining-for-marketing-and-business-intelligence/ThealgorithmforinducingarandomforestwasdevelopedbyLeoBreiman2001andAdeleCutler,and"RandomForests"istheirtrademarkThetermcamef

4、romrandomdecisionforeststhatwasfirstproposedbyTinKamHoofBellLabsin1995ThemethodcombinesBreiman's"bagging"ideaandtherandomselectionoffeatures,introducedindependentlybyHo1995andAmitandGeman1997inordertoconstructacollectionofdecisiontreeswithcontrolledvariation.4/25第五届中国R语

5、言会议北京2012李欣海随机森林简介Treemodelsyβ+βx+βx+βx+εi011i22i33iiClassificationtreeRegressiontreeCrawley2007TheRBookp691Crawley2007TheRBookp6945/25第五届中国R语言会议北京2012李欣海随机森林简介Thestatisticalcommunityusesirrelevanttheory,questionableconclusions?DavidR.CoxEmanuelParzenBruceHoadleyBradEfr

6、onNOYES6/25第五届中国R语言会议北京2012李欣海随机森林简介Ensembleclassifiers////.Treemodelsaresimple,oftenproducenoisybushyorweakstuntedclassifiersBaggingBreiman,1996:Fitmanylargetreestobootstrap-resampledversionsofthetrainingdata,andclassifybymajorityvoteBoostingFreund&Shapire,1996:Fitmany

7、largeorsmalltreestoreweightedversionsofthetrainingdata.ClassifybyweightedmajorityvoteRandomForestsBreiman1999:Fancierversionofbagging.IngeneralBoostingRandomForestsBaggingSingleTreeTrevorHastie.7/25第五届中国R语言会议北京2012李欣海随机森林简介HowRandomForestWorks////.Ateachtreesplit,arando

8、msampleofmfeaturesisdrawn,andonlythosemfeaturesareconsideredforsplittingTypicallymsqrtporlogp,wherepisthenumbe