传统电脑决策

TheApplicationofFuzzySetsonDataMiningProf.Tzung-PeiHongDepartmentofElectricalEngineeringNationalUniversityofKaohsiungT.P.Hong2OutlineIntroductionReviewDataMiningFuzzySetsFuzzyDataMiningFuzzyAssociationRules(I)FuzzyAssociationRules(II)FuzzyGeneralizedAssociationRulesFuzzyWebMiningT.P.Hong3OutlineReviewofGAFuzzydataminingformembershipfunctionsandrulesTwoapproachesConclusionT.P.Hong4ReasonsfordataminingSamGoodsSupermarketHowtoarrangegoodsintosupermarket?T.P.Hong5ReasonsfordataminingSamcustomerhowtomakethemarketingstrategiesforthesecustomer?T.P.Hong6MiningassociationrulesBreadMilkIFbreadisboughtthenmilkisboughtT.P.Hong7TheroleofdataminingUsefulpatternsTransactiondataPreprocessdataDataMiningKnowledgeandstrategyDatainformationT.P.Hong8DifferentkindsofknowledgeAssociationrulesGeneralizedassociationrulesSequentialpatternsQuantitativeassociationrulesClassificationrulesClusteringrulesetc…FocusT.P.Hong9MiningassociationrulesBreadMilkIFbreadisboughtthenmilkisboughtT.P.Hong10AprioriAlgorithmProposedbyAgrawaletal.Step1:Defineminsupandminconfex:minsup=50%minconf=50%Step2:FindlargeitemsetsStep3:GenerateassociationrulesT.P.Hong11ExampleLargeitemsetsTIDItems100ACD200BCE300ABCE400BEDatabaseC2Itemset{AB}{AC}{AE}{BC}{BE}{CE}C3Itemset{BCE}ScanDatabaseScanDatabaseScanDatabaseItemsetSup.{A}2{B}3{C}3{D}1{E}3C1ItemsetSup.{AB}1{AC}2{AE}1{BC}2{BE}3{CE}2C2ItemsetSup.{BCE}2C3ItemsetSup.{A}2{B}3{C}3{E}3L1ItemsetSup.{AC}2{BC}2{BE}3{CE}2L2ItemsetSup.{BCE}2L3T.P.Hong12ExampleAssociationrulesConfidenceIFBCTHENES(BCE)/S(BC)=2/2IFBETHENCS(BCE)/S(BE)=2/3IFCETHENBS(BCE)/S(CE)=2/2IFBTHENCES(BCE)/S(B)=2/3IFCTHENBES(BCE)/S(C)=2/3IFETHENBCS(BCE)/S(E)=2/3IFATHENCS(AC)/S(A)=2/2IFCTHENAS(AC)/S(C)=2/3IFBTHENCS(BC)/S(B)=2/3IFCTHENBS(BC)/S(C)=2/3IFBTHENES(BE)/S(B)=3/3IFETHENBS(BE)/S(E)=3/3IFCTHENES(CE)/S(C)=2/3IFETHENCS(CE)/S(E)=2/3T.P.Hong13FuzzySets傳統電腦決策不是對(1)就是錯(0)例如：25歲以上是青年，那26歲就是中年?60分以上是及格，那60分以下就是不及格何謂模糊在對(1)與錯(0)之間，再多加幾個等級幾乎對(0.8)可能對(0.6)可能錯(0.4)幾乎錯(0.2)T.P.Hong14FuzzySetsQuestion:168公分到底算不算高?身高(Cm)中矮高170180160隸屬度再多分成幾級連續T.P.Hong15Example:“Closeto0”e.g.μA(3)=0.01μA(1)=0.09μA(0.25)=0.62μA(0)=1DefineaMembershipFunction:μA(x)=2x1011T.P.Hong16Example:“Closeto0”VeryCloseto0:μA(x)=22)x1011(T.P.Hong17FuzzySet(Cont.)Membershipfunction[0,1]e.g.sunny:x→[0,1]0.6sunny0.8sunny0.1sunnyxT.P.Hong18FuzzySetSimpleIntuitivelypleasingAgeneralizationofcrispsetVaguemember→non-memberSunnyNotsunny10.80.60.40.200or1Non-membermembergradualT.P.Hong19FuzzyOperations交集(AND)取較小的可能性EX:學生聰明(0.8)而且用功(0.6)則是模範生(0.6)聯集(OR)取較大的可能性EX:學生聰明(0.8)或者用功(0.6)則是模範生(0.8)反面(NOT)取與1的差EX:學生聰明是0.8,則學生不聰明0.2T.P.Hong20FuzzyInferenceExample洪老師找小老婆的條件(大眼睛而且小嘴巴)或者是身材好Question:誰是最佳女主角大眼睛小嘴巴身材好陶晶瑩00.80.3張惠妹10.60.8李玟00.30.9李心潔0.70.10.5蔡依林0.80.50.3T.P.Hong21Answer對陶晶瑩=(0AND0.8)OR0.3=0OR0.3=0.3對張惠妹=(1AND0.6)OR0.8=0.8對李玟=(0AND0.3)OR0.9=0.9對李心潔=(0.7AND0.1)OR0.5=0.5對蔡依林=(0.8AND0.5)OR0.3=0.5李玟為最佳選擇!謝謝!T.P.Hong22FuzzyDecisionA={A1,A2,A3,A4,A5}AsetofalternativesC={C1,C2,C3}AsetofcriteriaC1(bigeyes)C2(smallmouth)C3(goodshape)A1(Mary)00.80.3A2(Judy)10.60.8A3(Jan)00.30.9A4(Mandy)0.70.10.5A5(Nancy)0.80.50.3T.P.Hong23Example(Cont.)Assume:C1andC2orC3E(Ai):evaluationfunctionE(A1)=(00.8)0.3=00.3=0.3E(A2)=(10.6)0.8=0.60.8=0.8E(A3)=(00.3)0.9=00.9=0.9thebestchoiceE(A4)=(0.70.1)0.5=0.10.5=0.5E(A5)=(0.80.5)0.3=0.50.3=0.5C1(bigeyes)C2(smallmouth)C3(goodshape)A1(Mary)00.80.3A2(Judy)10.60.8A3(Jan)00.30.9A4(Mandy)0.70.10.5A5(Nancy)0.80.50.3T.P.Hong24MotivationforFuzzyMiningInreal-worldapplicationsTransactionswithquantitativevaluesUsingfuzzysetstoprocessitTIDPurchaseditems1(A,3)(C,4)(E,2)2(B,3)(C,7)(D,7)3(B,2)(C,10)(E,5)4(A,9)(E,10)5(A,7)(D,8)6(B,2)(C,8)(D,10)T.P.Hong25FuzzyDataMiningSolvingquantitativevaluese.g.Johnbuys10bread,2butterand3Milk.FuzzyDataMiningMuchBreadandLittleButterMiddleAmountofMilkT.P.Hong26FuzzydataminingQuantitativedataLinguistictermMembershipfunction01611LowMiddleHigh10MembershipvalueNumberofitemT.P.Hong27MainIdeaNumericDatabaseFuzzyDataMiningKnowledge61218LowMiddleHighQuantity0NumericDatabasemilkbreadcookiesbeverageT14222T27307T38539T491513Ifmilk.MiddleThencookies.LowT.P.Hong28RelatedresearchLeeandHyung,1997α-cutConvertingthemembershiptuplestobinarytuplesPedrycz,1996,1998RunningtheFCM(FuzzyC-Means)methodSolvingclusterproblemT.P.Hong29RelatedresearchChanandhisco-workers,1997Maddourietal.,1998Rubin,1998Hanetal.,1998UsingMachinelearningmethodCombiningdataminingT.P.Hong30FuzzyMiningAlgorithmInputnquantitativetransactiondatamattributesAsetofmembershipfunctionsTwothresholdsMinimumsupport=Minimumconfidence=OutputAsetoffuzzyassociationrulesT.P.Hong31FuzzyMiningAlgorithmStep1Transformthequantitativevalueofeachtransactiondatumintoafuzzysetusingthegivenmembershipfunctions.Step2CalculatethescalarcardinalityofeachattributeregioninthetransactiondataT.P.Hong32FuzzyMiningAlgorithmStep3Foreachfuzzyregion,checkwhetheritisinthesetoflarge1-itemsets(L1)Step4Setr=1,whererisusedtorepresentthenumberofitemskeptinthecurrentlargeitemsetsT.P.Hong33FuzzyMiningAlgorithmStep5GeneratethecandidatesetCr+1fromLrinaw