【MOOC】交通數(shù)據(jù)挖掘技術(shù)（Data Mining for Transportation）-東南大學(xué) 中國大學(xué)慕課MOOC答案

【MOOC】交通數(shù)據(jù)挖掘技術(shù)（DataMiningforTransportation）-東南大學(xué)中國大學(xué)慕課MOOC答案Test11、【單選題】WhichoneisnotthedescriptionofDatamining?本題答案：【Appropriatestatisticalanalysismethodstoanalyzethedatacollected】2、【單選題】Whichonedescribestherightprocessofknowledgediscovery?本題答案：【Selection-Preprocessing-Transformation-Datamining-Interpretation/Evaluation】3、【單選題】WhichoneisnotbelongtotheprocessofKDD?本題答案：【Datadescription】4、【單選題】Whichoneisnottherightalternativenameofdatamining?本題答案：【Dataharvesting】5、【單選題】Whichoneisnotthenominalvariables?本題答案：【Age】6、【單選題】Whichoneiswrongaboutclassificationandregression?本題答案：【W(wǎng)ecanconstructclassificationmodels(functions)withoutsometrainingexamples.】7、【單選題】Whichoneiswrongaboutclusteringandoutliers?本題答案：【Clusteringbelongstosupervisedlearning.】8、【單選題】Aboutdataprocess,whichoneiswrong?本題答案：【W(wǎng)henmakingdataclassification,wepredictcategoricallabelsexcludingunorderedone.】9、【判斷題】Outlierminingsuchasdensitybasedmethodbelongstosupervisedlearning.本題答案：【錯誤】10、【判斷題】Supportvectormachinescanbeusedforclassificationandregression.本題答案：【正確】Test21、【單選題】Whichisnotthereasonweneedtopreprocessthedata?本題答案：【tomakeresultmeetourhypothesis】2、【單選題】Whichisnotthemajortasksindatapreprocessing?本題答案：【Transition】3、【單選題】HowtoconstructnewfeaturespacebyPCA?本題答案：【NewfeaturespacebyPCAisconstructedbyeliminatingtheweakcomponentstoreducethesizeofthedata.】4、【單選題】Whichoneiswrongaboutmethodsfordiscretization?本題答案：【Clusteringanalysisonlybelongstotop-downsplit.】5、【單選題】WhichoneiswrongaboutEqual-width(distance)partitioningandEqual-depth(frequency)partitioning?本題答案：【Theintervaloftheformeroneisnotequal.】6、【單選題】Whichoneiswrongwaytonormalizedata?本題答案：【Simplescaling】7、【多選題】Whicharetherightwaytofillinmissingvalues?本題答案：【Smartmean#Probablevalue#Ignore】8、【多選題】Whicharetherightwaytohandlenoisedata?本題答案：【Regression#Cluster#WT#Manual】9、【多選題】Whichoneisrightaboutwavelettransforms?本題答案：【TheDWTdecomposeseachsegmentoftimeseriesviathesuccessiveuseoflow-passandhigh-passfilteringatappropriatelevels.#Wavelettransformscanbeusedforreducingdataandsmoothingdata.】10、【多選題】Whicharethecommonusedwaystosampling?本題答案：【Simplerandomsamplewithoutreplacement#Simplerandomsamplewithreplacement#Stratifiedsample#Clustersample】11、【判斷題】Discretizationmeansdividingtherangeofacontinuousattributeintointervals.本題答案：【正確】Test31、【單選題】What'sthedifferencebetweeneagerlearnerandlazylearner?本題答案：【Eagerlearnerswouldgenerateamodelforclassificationwhilelazylearnerwouldnot.】2、【多選題】HowtochoosetheoptimalvalueforK?本題答案：【Cross-validationcanbeusedtodetermineagoodvaluebyusinganindependentdatasettovalidatetheKvalues.#LowvaluesforK(likek=1ork=2)canbenoisyandsubjecttotheeffectofoutliers.#Historically,theoptimalKformostdatasetshasbeenbetween3-10.】3、【多選題】What’sthemajorcomponentsinKNN?本題答案：【Howtomeasuresimilarity?#Howtochoosek?#Howareclasslabelsassigned?】4、【多選題】WhichoneofthefollowingwayscanbeusedtoobtainattributeweightforAttribute-WeightedKNN?本題答案：【Priorknowledge/experience.#PCA,FA(Factoranalysismethod).#Informationgain.#Gradientdescent,simplexmethodsandgeneticalgorithm.】5、【判斷題】AtlearningstageKNNwouldfindtheKclosestneighborsandthendecideclassifyKidentifiednearestlabel.本題答案：【錯誤】6、【判斷題】AtclassificationstageKNNwouldstoreallinstanceorsometypicalofthem.本題答案：【錯誤】7、【判斷題】Normalizingthedatacansolvetheproblemthatdifferentattributeshavedifferentvalueranges.本題答案：【正確】8、【判斷題】ByEuclideandistanceorManhattandistance,wecancalculatethedistancebetweentwoinstances.本題答案：【正確】9、【判斷題】DatanormalizationbeforeMeasureDistancecanavoiderrorscausedbydifferentdimensions,self-variations,orlargenumericaldifferences.本題答案：【正確】10、【判斷題】Thewaytoobtaintheregressionforanewinstancefromtheknearestneighborsistocalculatetheaveragevalueofkneighbors.本題答案：【正確】11、【判斷題】Thewaytoobtaintheclassificationforanewinstancefromtheknearestneighborsistocalculatethemajorityclassofkneighbors.本題答案：【正確】12、【判斷題】ThewaytoobtaininstanceweightforDistance-WeightedKNNistocalculatethereciprocalofthedistancesquaredbetweenobjectandneighbors.本題答案：【正確】Test41、【多選題】Whichdescriptionisrightaboutnodesindecisiontree?本題答案：【Internalnodestestthevalueofparticularfeatures#Leafnodesspecifytheclass】2、【多選題】ComputinginformationgainforcontinuousvalueattributewhenusingID3consistsofthefollowingprocedure:本題答案：【SortthevalueAinincreasingorder.#Considerthemidpointbetweeneachpairofadjacentvaluesasapossiblesplitpoint.#Selecttheminimumexpectedinformationrequirementasthesplit-point.#Split.】3、【多選題】Whichisthetypicalalgorithmstogeneratetrees?本題答案：【ID3#C4.5#CART】4、【多選題】Whichoneisrightaboutunderfittingandoverfitting?本題答案：【Underfittingmeanspooraccuracybothfortrainingdataandunseensamples.#Overfittingmeanshighaccuracyfortrainingdatabutpooraccuracyforunseensamples.#Underfittingimpliesthemodelistoosimplethatweneedtoincreasethemodelcomplexity.#Overfittingoccurstoomanybranchesthatweneedtodecreasethemodelcomplexity.】5、【多選題】Whichoneisrightaboutpre-pruningandpost-pruning?本題答案：【Bothofthemaremethodstodealwithoverfittingproblem.#Pre-pruningdoesnotsplitanodeifthiswouldresultinthegoodnessmeasurefallingbelowathreshold.#Post-pruningremovesbranchesfroma“fullygrown”tree.】6、【多選題】Post-pruninginCARTconsistsofthefollowingprocedure:本題答案：【First,considerthecostcomplexityofatree.#Then,foreachinternalnode,N,computethecostcomplexityofthesubtreeatN.#AndalsocomputethecostcomplexityofthesubtreeatNifitweretobepruned.#Atlast,comparethetwovalues.IfpruningthesubtreeatnodeNwouldresultinasmallercostcomplexity,thesubtreeispruned.Otherwise,thesubtreeiskept.】7、【判斷題】ThecostcomplexitypruningalgorithmusedinCARTevaluatecostcomplexitybythenumberofleavesinthetree,andtheerrorrate.本題答案：【正確】8、【判斷題】GainratioisusedasattributeselectionmeasureinC4.5andtheformulaisGainRatio(A)=Gain(A)/SplitInfo(A).本題答案：【正確】9、【判斷題】Ruleiscreatedforeachpartfromitsroottoitsleafnotes.本題答案：【正確】10、【判斷題】ID3useinformationgainasitsattributeselectionmeasure.AndtheattributewiththelowestinformationgainischosenasthesplittingattributefornoteN.本題答案：【錯誤】Test51、【多選題】WhatthefeatureofSVM?本題答案：【Extremelyslow,butarehighlyaccurate.#Muchlesspronetooverfittingthanothermethods.#Provideacompactdescriptionofthelearnedmodel.】2、【多選題】Whichisthetypicalcommonkernel?本題答案：【Linear#Polynomial#Radialbasisfunction(Gaussiankernel)#Sigmoidkernel】3、【多選題】WhatadaptationscanbemadetoallowSVMtodealwithMulticlassClassificationproblem?本題答案：【Oneversusrest(OVR).#Oneversusone(OVO).#Errorcorrectingoutputcodes(ECOC).】4、【多選題】What'stheproblemofOVR?本題答案：【Sensitivetotheaccuracyoftheconfidencefiguresproducedbytheclassifiers.#Thescaleoftheconfidencevaluesmaydifferbetweenthebinaryclassifiers.#Thebinaryclassificationlearnersseeunbalanceddistributions.】5、【多選題】WhichoneisrightabouttheadvantagesofSVM?本題答案：【Theyareaccurateinhigh-dimensionalspaces.#Theyarememoryefficient.#Thealgorithmisnotproneforover-fittingcomparedtootherclassificationmethod.#Thesupportvectorsaretheessentialorcriticaltrainingtuples.】6、【判斷題】Kerneltrickwasusedtoavoidcostlycomputationanddealwithmappingproblems.本題答案：【正確】7、【判斷題】ThereisnostructuredwayandnogoldenrulesforsettingtheparametersinSVM.本題答案：【正確】8、【判斷題】Errorcorrectingoutputcodes(ECOC)isakindofproblemtransformationtechniques.本題答案：【錯誤】9、【判斷題】Regressionformulasincludingthreetypes:linear,nonlinearandgeneralform.本題答案：【正確】10、【判斷題】Ifyouhaveabigdataset,SVMissuitableforefficientcomputation.本題答案：【錯誤】Test61、【多選題】Whichdescriptionisrighttodescribeoutliers?本題答案：【Outlierscausedbymeasurementerror#Outliersreflectinggroundtruth#Outlierscausedbyequipmentfailure】2、【多選題】Whatisapplicationcaseofoutliermining?本題答案：【Trafficincidentdetection#Creditcardfrauddetection#Networkintrusiondetection#Medicalanalysis】3、【多選題】Whichoneisthemethodtodetectoutliers?本題答案：【Statistics-basedapproach#Distance-basedapproach#Density-basedapproach】4、【多選題】Howtopicktherightkbyaheuristicmethodfordensity-basedoutlierminingmethod?本題答案：【Kshouldbeatleast10toremoveunwantedstatisticalfluctuations.#Pick10to20appearstoworkwellingeneral.#Picktheupperboundvalueforkasthemaximumof“closeby”objectsthatcanpotentiallybelocaloutliers.】5、【多選題】Whichoneisrightaboutthreemethodsofoutliermining?本題答案：【Statistics-basedapproachissimpleandfastbutdifficulttodealwithperiodicitydataandcategoricaldata.#Theefficiencyofdistance-basedapproachislowforthegreatdatasetinhighdimensionalspace.】6、【判斷題】Distance-basedoutlierMiningisnotsuitabletodatasetthatdoesnotfitanystandarddistributionmodel.本題答案：【錯誤】7、【判斷題】Statistic-basedmethodneedstorequireknowingthedistributionofthedataandthedistributionparametersinadvance.本題答案：【正確】8、【判斷題】Whenidentifyingoutlierswithadiscordancytest,thedatapointisconsideredasanoutlierifitfallswithintheconfidenceinterval.本題答案：【錯誤】9、【判斷題】MahalanobisDistanceaccountsfortherelativedispersionsandinherentcorrelationsamongvectorelements,whichisdifferentfromEuclideanDistance.本題答案：【正確】10、【判斷題】Anoutlierisadataobjectthatdeviatessignificantlyfromtherestoftheobjects,asifitweregeneratedbyadifferentmechanism.本題答案：【正確】Test71、【多選題】Howtodealwithimbalanceddatain2-classclassification?本題答案：【Oversampling#Undersampling#Threshold-moving#Ensembletechniques】2、【多選題】Whichoneisrightwhendealingwiththeclass-imbalanceproblem？本題答案：【Smotealgorithmaddssynthetictuplesthatareclosetotheminoritytuplesintuplespace.#Threshold-movingandensemblemethodswereempiricallyobservedtooutperformoversamplingandundersampling.】3、【多選題】Whichstepisnecessarywhenconstructinganensemblemodel?本題答案：【Creatingmultipledataset#Constructingasetofclassifiersfromthetrainingdata#Combiningpredictionsmadebymultipleclassifierstoobtainfinalclasslabel】4、【判斷題】Ensemblestendtoyieldbetterresultswhenthereisasignificantdiversityamongthebasemodels.本題答案：【正確】5、【判斷題】EnsemblemethodcannotparallelizablebecausenoteverybaseclassifiercanbeallocatedtoadifferentCPU.本題答案：【錯誤】6、【判斷題】Togeneratethesingleclassifier,differentmodelmaybeusedtodealwithdifferentdatasubset.本題答案：【正確】7、【判斷題】Inrandomforest,usingar