决策树分类器算法的研究

《决策树分类器算法的研究》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、北京交通大学硕士学位论文决策树分类器算法的研究姓名：黄泽宇申请学位级别：硕士专业：计算机应用技术指导教师：王志海20060301北京交通大学硕士学位论文AbstractClassificationisoneofveryimportantbasictasksinmefieldofdatamiIlingandmachjneleaming．ItcanbeusedtoaIlalyzea11dstudyavastnumberofrelateddataandestablishcl嬲si聊ngmodeIsinmanya

2、reasofrelatedproblems．Theclassi矗cationtechlliqucshaveextensive印plicationusageinscientificresearch，commuIlication，finallceandothernelds．Adecision．ⅡeeclassmerisaveryimportaIltmodelintlleprocessofhowledgediscoVeⅨGoodimerpretabili劬f船tclassificationspeeda11deX

3、cellcntclassificationperfbrmanceofdecision-仃eemakeit晷aduallybecomemeresearchfocusinthefieldsofdatamillingaIldmachineleamin吕Themostclassicaldecision—n℃ele蛐[1ingsystemisD3，whichuse也edivide-aIld-conquer印pmachtodecision一仃它einductionfromroottoleaves，a11dchoose

4、mespli牡ingat仃曲utesbytheg咖ratio．ThismemOdcanensurctocons眦tasimple仃ee．Butm3caIlnoth觚dlen啪ericattributes，onlynominala船讪utes．nisusu砒lyover虹ttedtome仃ainingdatabases．C4．5algorimmistlleextensionofD3．Itextendsmeclassificationability0fID3疗omnominalattributestonume

5、ricattributes．Itwellres01vesthepmblemaboutov础仕ingbypnlningdecision·trees．Nowithasahadybeenhownasabe戗erdecision-仃ceclassi丘er．Lazydecision·打eesalsoco璐虮lctdecision-trees，adoptl配yalgorithm，a11ddelaymele锄ingproceedinguntilatestinstaIlceis百Ven．Itconc印tllaIlycon

6、s缸uctsⅡle‘‘best”decision一仃eeforeachtestinstance．h1smalldat如ases，itsclassificationaccllracyisVery11i曲．Butinsomelargerdatabases，itsslowcl船smcationspeed，largestorecostingaIldeasilyinducedbynoisesa丘bcti招cl嬲sificationperfo舰aIlce，speciallyinme1a曙erd砒lbasesteste

7、dwllichhaveagreatdealof砒rfbutes．Basedon仕屺allalysisofD3algo删撇，C4．5algOrithIn，1azydecision-tr∞sclassifiersalgori伽n，andNaiveBayesclassifiersalgorithrn，t11reenewdecision-treeclassifiersarepresemedinmispaper'whichareadecision—treeclassificrhybridmodelofeagers仃

8、．negya11dlazys打ategy，aⅡ摘要decision—treeclassmermodelofcompetitivelyselectingsplittingattributes，aIlda1azydecision．treeclassi壬ieronbaseofwci星manddistallce．Inmisp印er，wealsocomparethenewmodelswithD3algorithrn，C4．5algori