Integrating multiple learned classification models (classifiers) computed over large and (physically) distributed data sets has been demonstrated as an e#ective approach to scaling inductive learning techniques, while also boosting the accuracy of individu
MinimalCostComplexityPruningofMeta-Classi ers
AndreasL.ProdromidisandSalvatoreJ.StolfoDepartmentofComputerScienceColumbiaUniversityNewYork,NY10027{andreas,sal}@cs.columbia.edu
Integratingmultiplelearnedclassi cationmodels(classi ers)computedoverlargeand(physically)dis-tributeddatasetshasbeendemonstratedasane ec-tiveapproachtoscalinginductivelearningtechniques,whilealsoboostingtheaccuracyofindividualclassi- ers.Thesegains,however,comeattheexpenseofanincreaseddemandforrun-timesystemresources.The nalensemblemeta-classi ermayconsistofalargecollectionofbaseclassi ersthatrequireincreasedmemoryresourceswhilealsoslowingdownclassi ca-tionthroughput.Toclassifyunlabeledinstances,pre-dictionsneedtobegeneratedfromallbase-classi ersbeforethemeta-classi ercanproduceits nalclassi- cation.Thethroughput(predictionrate)ofameta-classi erisofsigni cantimportanceinreal-timesys-tems,suchasine-commerceorintrusiondetection.
Thisextendedabstractdescribesapruningalgorithmthatisindependentofthecombiningschemeandisusedfordiscardingredundantclassi erswithoutdegradingtheoverallpredictiveperformanceoftheprunedmeta-classi er.Todeterminethemoste ectivebaseclassi- ers,thealgorithmtakesadvantageoftheminimalcost-complexitypruningmethodoftheCARTlearningalgo-
rithm(Breimanetal.1984)whichguaranteesto ndthebest(withrespecttomisclassi cationcost)prunedtreeofaspeci csize(numberofterminalnodes)ofaninitialunpruneddecisiontree.AnalternativepruningmethodusingRissanen’sminimumdescriptionlengthisdescribedin(Quinlan&Rivest1989).
Minimalcostcomplexitypruningassociatesacom-plexityparameterwiththenumberofterminalnodesofadecisiontree.Itprunesdecisiontreesbyminimiz-ingthelinearcombinationofthecomplexity(size)ofthetreeanditsmisclassi cationcostestimate(errorrate).Thedegreeofpruningiscontrolledbyadjustingtheweightofthecomplexityparameter,i.e.anincreaseofthisweightparameterresultsinheavierpruning.
Pruninganarbitrarymeta-classi erconsistsofthreestages.Firstweconstructadecisiontreemodel(e.g.CART)oftheoriginalmeta-classi er,bylearningitsinput/outputbehavior.Thisnewmodel(adecision
Minimal cost complexity ... 1页 免费 Analysis of a complexity... 17页... Cost Complexity-based Pruning of Ensemble Classi?ers? Andreas L. ...
Analysis of a complexity based pruning scheme for classification trees A complexity based pruning procedure for classication trees is described, and bounds ...
Minimal Cost-Complexity Pruning 未剪枝前的树记为 Tmax ,剪枝过程包括两步: ①按照某种规则,得到树 Tmax 的参数族 (候选剪枝树) Tmax ,T1 , T2 ,…,{t...
The new method has an order of computational complexity comparable to cost-...Key words: decision trees; tree pruning; data mining; classi?cation History...
Minimal complexity attack classification intrusion detection system_法律资料_人文社科_专业资料。最小的复杂攻击分类入侵检测系统 摘要 在一般情况下, 用户和注入...
主要策略 : l) MCCP (MinimalCost一Complexitypruning) 2) PEP(PessimisticErrorPruning) 3) MEP(MinimumErrorPruning) 4) REP(ReducedErrorPunring) 5) MDL(...
? 常见的六种修剪技术: 1.MCCP(Minimal cost-complexity Pruning)-用 于CRAT系统。 2.REP(Reduced Error Pruning) 。 3.MEP(Minimal Error Pruning) 。 4....
The search complexity is given in terms of leaf-node evaluations (hence ... since it used additional special-purpose pruning methods, it supports the ...
最小代价复杂性剪枝法(Minimal Cost Complexity Pruning,MCCP) ? 目标:得到恰当的树,具有一定的预测精度, 且复杂程度恰当 ? 代价(误差)和复杂度之间的权衡是必要...
108 109 111 113 116 116 Chapter 6 A-posteriori Pruning of Meta-Classi?...ers. . . . The six steps of the Cost Complexity-based Post-Training ...
热门文档