分支合并對決策樹歸納學習的影響.pdf

1、河北大學碩士學位論文分支合并對決策樹歸納學習的影響姓名：楊晨曉申請學位級別：碩士專業(yè)：計算機應用技術指導教師：王熙照20070601AbstraclAbstractDecisiontreeisoneofthemostwidelyusedalgorithmsincommercialdataminingtoolsAndwenameitdecisiontreebecauseitiSatreeandisusedtodosomedecisions

2、It’Samethodforapproximatingdiscretevaluedtargetfunctions，inwhichthelearnedfunctionisrepresentedbyadecisiontreeThismodelcandenotethepatternofdescribingclassification，andbeusedtoaccomplishpredictiontasksAtpresentdecisiontr

3、eeisthemaintechniqueinthepredictionandclassification，especiallyincommercialfields，whichhasbeensuccessfullyappliedtoabroadrangeoftasksfromlearningtodiagnosemedicalcasetolearningtoassesscreditriskofloanapplicantsInshort，de

4、cisiontreecanhelpyoutransformdataintoknowledgeSinceinductivebiasexistduringtheprocessofselectionofexpandedattributes，attributewithmorevaluesareusuallypreferredtobeselectedItconsequentlyresultisadecisiontreewithlargescale

5、andwithpoorgeneralizationcapabilityThereforeitisnecessarytosimplifythedecisiontreeincludingpre—pruningandpostpruningTiffspaperwillfocusonthepre—prtmingAne訂strategy(mergingbranches)ofpre。pruningisgiven，thatis，attheprocess

6、oftreegrowth，twobranches(ormore)fromthesamenodemergeintoonebranchandthenthetreegrowthprocesscontinuesFinally，thereisverygreatdisparitybetweenthetreeswithandwithoutmerginginaspectofthestructure，size，complexityandtheaccura

7、cyWeproposedtwoalgorithmsofmergingbranchesfirstly：SSID(basedontheproportionofpositivesamples)andMID(basedontheMARGINEnlargementofSVM)Andthenbasedontheinformationgain，thispapertheoreticallyanalyzesthecomplexityofadecision

8、treebeforeandaftermergingbranches，anddesignsonealgorithmofmergingbranchesMCID(basedonthemostgaincompensation)Experimentalresultsshowthatwithrespecttothecomprehensibilityandthegeneralizationcapabilitythethreealgorithmsare