weka使用報(bào)告

1、WEKA使用實(shí)驗(yàn)報(bào)告一、實(shí)驗(yàn)?zāi)康臄?shù)據(jù)挖掘是通過分析存在于數(shù)據(jù)庫(kù)里的數(shù)據(jù)來解決問題。在數(shù)據(jù)挖掘中計(jì)算機(jī)以電子化的形式存儲(chǔ)數(shù)據(jù)，并且能自動(dòng)的查詢數(shù)據(jù)，通過關(guān)聯(lián)規(guī)則、分類與回歸、聚類分析等算法對(duì)數(shù)據(jù)進(jìn)行一系列的處理，尋找和描述數(shù)據(jù)里的結(jié)構(gòu)模式，進(jìn)而挖掘出潛在有用的信息。WEKA是一種開源的數(shù)據(jù)挖掘工具。WEKA的全名是懷卡托智能分析環(huán)境(WaikatoEnvironmentforKnowledgeAnalysis)，是-款免費(fèi)的，非商業(yè)化的數(shù)據(jù)挖掘工具，其源代碼可從( HYPERLINK http:/www.cs.waikato.acm http:/www.cs.waikato.acmz./ml/w

2、eka/)得到，我們?cè)诒敬螌?shí)驗(yàn)中所使用到的相關(guān)數(shù)據(jù)，也是從該處獲得的。WEKA作為一個(gè)公開的數(shù)據(jù)挖掘工作平臺(tái)，集合了大量能承擔(dān)數(shù)據(jù)挖掘任務(wù)的機(jī)器學(xué)習(xí)算法，包括對(duì)數(shù)據(jù)進(jìn)行預(yù)處理，分類，回歸，聚類，關(guān)聯(lián)規(guī)則以及在新的交互式界面上的可視化。本次試驗(yàn)，我們要通過學(xué)習(xí)WEKA工具的使用，與上課內(nèi)容相結(jié)合，針對(duì)某些數(shù)據(jù)挖掘算法建立起數(shù)據(jù)挖掘模型，進(jìn)而對(duì)數(shù)據(jù)分析技術(shù)有更深層次的了解。二、實(shí)驗(yàn)準(zhǔn)備在啟動(dòng)WEKA時(shí)，會(huì)彈出GUI選擇器，選擇使用WEKA和數(shù)據(jù)的四種方式。如下圖所示:GwckftCUEChwsflfj_ToolsclpAppLicataaits.WEKAEkpLnrMTheUmverstlynbV

3、iTG-nBIHtZrF輝4Clpj沁.FWTtEhoiflity5n5EEhtfTlLBWdWS-ltyZJSBT：ZHFGgdl-DA昭曲小1?11|耳xwtA“垃曠沢L380L的0gCo.c0.01D012222a22222333L9S*l_l,23EtOLW.D弓D.CD.aL.L1JLJa.anw?L.CflSSSC.HZSH3LK0QQO.C0.0g遜；OjTTTTTS(13333330W888Q420220D9CILO0.0305SEL5.2&3E.0BS&95LGL.OL39.DIQD.Cd.o0.KSM&a.LSIBSD.13L13LO.37ZL33323508809uO.C

4、0.0CitLllLl0畸飢(19*1445-&7G32D9:D.Oo.aO9444-M1D&2B4LT77731.3hD9251.530L97D9C0.00.0LILI07-12TIM4fl2.L3W5217.0曲1-J0.L3L11L01JL13L3LftM；30fl6?1.2l27t詢90L7iD$:些0.01&S512TTTTa1.百他kLIl-B0L17.D9CD.2ZZ2770.0O0333330?Ti5Wl.QO.MM3l2I.3S-0LuQCO.CCl.flG0lTTTS0717713O3074U7L321407909CD.C0.0L277TB0S29B3L333Z0EO333

5、3L-lLJ30L25.D9C0.00.0OTJ3335n0OHJdCTO.3442!T5l5055u0cO.Co.oC-5S555E045S4IVit72223O我Lg7(X03S.D生qD.Co.aL.22Z22a.272L66i.a0.LT豹0陀2口0.CO.Q151.277詣L4L13L2.2環(huán)LSlEL0LSdQCO.C0.0S：i5bE1415?D7222Z2o型riseLT98.0LK.Di.CD.Co.a.竺竺D.H3&5D.133333r.20皿00g-jB.C0.01:0陽(yáng)1.00.577n1.LE20237DgD.L1L11Lo.a22777B1143-S22fiK72.

6、5&472型LK.O2.Pi.CD.Ca.QQ.2K73-L.鬧Q.2?LHS23&30Lffi0Q口O.C0.02：38ES9L4W9L4SJ3303.OW052494:LEKD9C些o.aeetidii1丐i.a0516339曲230.0LUDi.CD.Cn.o2:irras?1.39-45a15?0啟09E.d0.01722221.34LS70222222QT2L3327IS0皿D9CILOo.aC-33SS3弓a01851DfilLUI.O950L9Dgc0.0013L11L3.LMflTZfl.O3.1.0545L12.038ugcO.C0.01.115bH0743L3L1JL：L4

7、6305za2K.0L33.DQ.CD.Oo.aO.KS&a.L364B3D.畫SSSO.】曲彌312EZ0L45D9C0.00.0OJJ0C0.00.0217.0遜QO.Co.o.-s2Io46?264b6：2曲皿Z3LB9.0LB；.DD.CD.a1.2232L.M372.DK0Z1220幻09uO.C0.01.esj0詢曲1)3SSi-0L30tigijO.C0.0G7S2220SD74D7L:SSSbG-H|f|Yh希|CCi&r-a三、實(shí)驗(yàn)內(nèi)容1.貝葉斯算法點(diǎn)“Choose”按鈕選擇“bayes”，這是WEKA中實(shí)現(xiàn)的貝葉斯算法。選擇Cross-Validatioinfolds=10

8、,然后點(diǎn)擊“start”按鈕，就可以生成貝葉斯模型并且該模型的誤差分析結(jié)果出現(xiàn)在右邊“Classifieroutput”中。下圖是本例中訓(xùn)練數(shù)據(jù)集的誤差分析結(jié)果：RodiDrormacionScJiHnewtJra,clnssirterfl,b3jTE3NiBLvta匹3EtelaClODf3E-3E&DZInstEnces:15D0ft.tcrltoiefl：2Dr-esLon-cezirrid-CDLrtgion-piKe1-c-dudt：shDiL-Liae-il-ejiBiT.y-5vr-Sge-rcanv-&3de-sd.h.-sdgereanir-3flr-adlncenslLy-

9、seaiLee.wred-meanZElflDLue-TTtaZlaHHareen-neajLexr-ed-raianeitbLue-TOanix-sreeD-neaiiVEloe-Mian3acurablo?i-iMa-nHoe-EteaDeLe.53Teaci5ode510-f&ldcraaa-va1ldaTlcmCla35j.fierUDdeL(fulltrai.口:Ling1set)=WaiveEay鹽Cla.ssLLezClasskinrlbuiebrlcklace卩-14|sky(a-is?CollageWSceirfiit0-15?niLDdDlrf0-14?p(iuh(0-1-

10、1gassegrelaD-ceDzrc44-CDLZIS-EDQ0.7EE7134.B37E91.3949129.d157-06212S.24Sa134.DBEEacdrEev-63r526769,2729TO,a93Efi7,2da56?r92017L.9334H.5L3Lwslhzmum20522Q22Q20i23S2D7TSE-ELSlOnLtH1.004-1.004.1.0041.0041.DD4-Srrar12zedcrDBS-ralldac丄lhiSuzmary12Lfi2540.77910.05540.225622.6144%64.5645*ai.07$LS-9333Corrc-

11、clyClassiridrngcanoesLnaarrscLLyClassilelIjisracsKappaBtatisciaMeanahs&lutserrorRootnreanaquGrrdsrrarRelariiTEatsol口匚己=rrarRc&cr-ElativeaquarederrorTc-talJlmibeioflastance1500=De匚日ilEdccuracyEyClBBS=TPRatsFPSawFreciji&nRecallE-JfeaajzeBDCAi=sCL33a0.S460,0450r75S0.546,0420r?67b匚iulcfia二匕a_99LD10.?910

12、.995LskOulSED*OL10.7360-.1SSm細(xì)0.S15SoliageCL酣4：D.0270.5440.554CL弱4o.g4GEment072LD.130.4ti50.721O.5C50.293vindjw0-945D.0D30.5S2D/45a.9630-999parK099D10.S9-9950.S99grassweightedAvg-0b1L0.03U.S32o.en0.7930.97=ConfusionBatrlx=hcdEf?-alasdifLedas194005s001a二brickfaaeD2ie0200D1b=5ky6396157a1c=toLlage17011

13、90401d-aern39c百121470D1E二windowD0490223D1f=pathan20205Ig=grassmwi同時(shí)，還可以得到校驗(yàn)數(shù)據(jù)集的誤差結(jié)果，如下圖所示:=Evalnailonont-earsec=5uircnarY=CorrectlyClassifiedInstances2417-.037ineorreDclyClassifie-dInacancesISfi22.9E3Kappacaziatic0.73LCMeanabsoluteerrur0.0659Roc-tmeanstn-arederror.Z464Relativesabsolute-error26.S912i

14、Roc-tre1azivesqjarederror70.34,31冬TotalEwnherofInatarsceaE10=EetoiL-edAccuzocyEyCLas3=TPRateFPRatePrecisionRecallF-MeasiJireROCAr-eaC13330.9520.0l0-7290.9520.532o.geebri亡rfaeeDu951010.9910.9950*999skyD.1390.0130S4CM330.23Om93Sfo-1iage0.6620.Q53o.eoe00020.6430-973匚二n亡且上D.5350.1590.40S0.5950.4340.55Ir

15、fllld&Wu.y北0,004a.9670.9360.9510.359pach0.96701D.Qfi?0.9230.994graaaWeightedivg*D.710-0410-.750*770*7130 x951=CanfuaiDDHat匚讓=己cde亠j-cLasslZle-1as1190015001a=brickfaz0109L10001b=sky3DIT1101001a-foliage901970301d二aeinmt3001575001e=window00150ee01二=parti0010301191a=grass本次校驗(yàn)數(shù)據(jù)集的準(zhǔn)確率為77.037%，準(zhǔn)確率不高。2.決策樹算

16、法訓(xùn)練數(shù)據(jù)集得到的決策樹為下圖所示:IgifcTrfrn筑&HiMtrsJiitCwkE.pIia?ci：vin4rinruwef.3irawsssrJi:&fe-i和咚I沖曲J-MTT7F耆注:護(hù)/0衲沖1A；brKITKi1.3iiMidIP沐H131WFp3DT*4TU!f*ld|J丫紳亍弧齊砂CHIT-iiMKnsTJii曲liWViR*1mnun茁，托怦門*打吃i*4WW2trEwtFHiy恃或ms謂-|14|忙|.PI.s.rid対v=13|-u二巧.。單“ncrrnriFH護(hù)CM呦段W*I叱D內(nèi)EMIW3W3ir同時(shí)我們還可以查看誤差分析:一三StzEtifirdc-iDaa-v

17、ElidaticnSumuar?Ccrrectly1133111.2a7勺KappascarlaGlc0.95D2Ite呂elatQluzeerror0.D13SRcotran耳e口曰匸皀zice-de-0L057ReLativeatsaluzeerror5647L%Rootrelative-aquuredtrz-or30,2115%ToCaLdumberofInatonceaIS00DetailelAccuracyfiyClasaIPRaceEPRawPMCiBlonRaeaLlE-N-Sdaur-eRDC型心ClassDuQ5O.OD40-575Du55fiD胡霹0.9S1brickfac

18、-E1O.OD10.9951o.se1=kyD.Q42o.oieG.E950.942D.9L60.975fnliag-eD.341O.OD90,945D,941D,3430,978ermrut0,9770,0170.B910,877DBB40,91window0.3B70.0010.99E0.9870.3910.99-7pathD.9Q1D購(gòu)Lgia.saWelgti-edS.vg.D.35T0.QD70.952D.357D.95T0985=匚口rLCualcnKatrlx=ahcdef9-匚Ifissifi-=das19fiD31500110-bricjtfacr：022000a001Ibw

19、jjty012300Ia*foliaa=20目207E_C1l1虧emen匸30179aC11e-WLEldCW00a3a233011f-parh00a020205|ig-gxass校驗(yàn)數(shù)據(jù)集后得到結(jié)果為:=Eval口日匚landelresEme匸=Ewnnai:/CorrtGtlyClassifi-cdTastiness巧9sfi.nze%IncD-rrectl/ClassifiedLnstiaces3.S272%XanpasrarLs匚1亡0.9553Mein,absoluteerror0.0127RoorEieansquaredrror0*1005Relati7tabaDluteerror5,1771aSootralarlvesmiarederror2S-.Q507tTottalNurkrofInstanceseio=DecaLiedAccuracyCZl己日m=I?RazeEFRarerCL3LOHRaealLF-Kea.siLMROCHraaClass0.S52D,D040.976D,9?20,9540,994brickface101L11sKyQ.975D.D190.902D.9750.937O.9S6foliage0.373D.D10.3390.9730.95S