大数据数据挖掘培训讲义7-回归和KNN算法.ppt

《大数据数据挖掘培训讲义7-回归和KNN算法.ppt》由会员上传分享，免费在线阅读，更多相关内容在PPT专区-天天文库。

1、ClassificationAlgorithms–ContinuedOutlineRulesLinearModels(Regression)Instance-based(Nearest-neighbor)2GeneratingRulesDecisiontreecanbeconvertedintoarulesetStraightforwardconversion:eachpathtotheleafbecomesarule–makesanoverlycomplexrulesetMoreeffectiveconversionsarenottrivi

2、al(e.g.C4.8testseachnodeinroot-leafpathtoseeifitcanbeeliminatedwithoutlossinaccuracy)3CoveringalgorithmsStrategyforgeneratingarulesetdirectly:foreachclassinturnfindrulesetthatcoversallinstancesinit(excludinginstancesnotintheclass)Thisapproachiscalledacoveringapproachbecausea

3、teachstagearuleisidentifiedthatcoverssomeoftheinstances4Example:generatingaruleIftruethenclass=a5Example:generatingarule,IIIfx>1.2thenclass=aIftruethenclass=a6Example:generatingarule,IIIIfx>1.2thenclass=aIfx>1.2andy>2.6thenclass=aIftruethenclass=a7Example:generatingarule,IVI

4、fx>1.2thenclass=aIfx>1.2andy>2.6thenclass=aIftruethenclass=aPossiblerulesetforclass“b”:Morerulescouldbeaddedfor“perfect”rulesetIfx1.2thenclass=bIfx>1.2andy2.6thenclass=b8Rulesvs.treesCorrespondingdecisiontree:(producesexactlythesamepredictions)But:rulesetscanbemoreclearwhe

5、ndecisiontreessufferfromreplicatedsubtreesAlso:inmulti-classsituations,coveringalgorithmconcentratesononeclassatatimewhereasdecisiontreelearnertakesallclassesintoaccount9AsimplecoveringalgorithmGeneratesarulebyaddingteststhatmaximizerule’saccuracySimilartosituationindecision

6、trees:problemofselectinganattributetosplitonBut:decisiontreeinducermaximizesoverallpurityEachnewtestreducesrule’scoverage:witten&eibe10SelectingatestGoal:maximizeaccuracyttotalnumberofinstancescoveredbyruleppositiveexamplesoftheclasscoveredbyrulet–pnumberoferrorsmadebyruleSe

7、lecttestthatmaximizestheratiop/tWearefinishedwhenp/t=1orthesetofinstancescan’tbesplitanyfurtherwitten&eibe11Example:contactlensdata,1Ruleweseek:Possibletests:Age=Young2/8Age=Pre-presbyopicAge=PresbyopicSpectacleprescription=MyopeSpectacleprescription=HypermetropeAstigmatism=

8、noAstigmatism=yesTearproductionrate=ReducedTearproductionrate=NormalIf?the