【答案】《交通數(shù)據(jù)挖掘技術(shù)(DataMiningforTransportation)》(東南大學)章節(jié)作業(yè)慕課答案

有些題目順序不一致，下載后按鍵盤ctrl+F進行搜索Week1.IntroductiontodataminingTest11.單選題：Aboutdataprocess,whichoneiswrong?

選項：

A、Whenmakingdatadiscrimination,wecomparethetargetclasswithoneorasetofcomparativeclasses(thecontrastingclasses).

B、Whenmakingdataclassification,wepredictcategoricallabelsexcludingunorderedone.

C、Whenmakingdatacharacterization,wesummarizethedataoftheclassunderstudy(thetargetclass)ingeneralterms.

D、Whenmakingdataclustering,wewouldgroupdatatoformnewcategories.

答案：【W(wǎng)henmakingdataclassification,wepredictcategoricallabelsexcludingunorderedone.】2.單選題：Whichoneiswrongaboutclusteringandoutliers?

選項：

A、Clusteringbelongstosupervisedlearning.

B、Principlesofclusteringincludemaximizingintra-classsimilarityandminimizinginterclasssimilarity.

C、Outlieranalysiscanbeusefulinfrauddetectionandrareeventsanalysis.

D、Outliermeansadataobjectthatdoesnotcomplywiththegeneralbehaviorofthedata.

答案：【Clusteringbelongstosupervisedlearning.】3.單選題：Whichoneiswrongaboutclassificationandregression?

選項：

A、Regressionanalysisisastatisticalmethodologythatismostoftenusedfornumericprediction.

B、Wecanconstructclassificationmodels(functions)withoutsometrainingexamples.

C、Classificationpredictscategorical(discrete,unordered)labels.

D、Regressionmodelspredictcontinuous-valuedfunctions.

答案：【W(wǎng)ecanconstructclassificationmodels(functions)withoutsometrainingexamples.】4.單選題：Whichoneisnotthenominalvariables?

選項：

A、Occupation

B、Education

C、Age

D、Color

答案：【Age】5.單選題：Whichoneisnottherightalternativenameofdatamining?

選項：

A、Knowledgeextraction

B、Dataarcheology

C、Datadredging

D、Dataharvesting

答案：【Dataharvesting】6.單選題：WhichoneisnotbelongtotheprocessofKDD?

選項：

A、Datamining

B、Datadescription

C、Datacleaning

D、Dataselection

答案：【Datadescription】7.單選題：Whichonedescribestherightprocessofknowledgediscovery?

選項：

A、Selection-Preprocessing-Transformation-Datamining-Interpretation/Evaluation

B、Preprocessing-Transformation-Datamining-Selection-Interpretation/Evaluation

C、Datamining-Selection-Interpretation/Evaluation-Preprocessing-Transformation

D、Transformation-Datamining-election-Preprocessing-Interpretation/Evaluation

答案：【Selection-Preprocessing-Transformation-Datamining-Interpretation/Evaluation】8.單選題：WhichoneisnotthedescriptionofDatamining?

選項：

A、Extractionofinterestingpatternsorknowledge

B、Explorationsandanalysisbyautomaticorsemi-automaticmeans

C、Discovermeaningfulpatternsfromlargequantitiesofdata

D、Appropriatestatisticalanalysismethodstoanalyzethedatacollected

答案：【Appropriatestatisticalanalysismethodstoanalyzethedatacollected】9.單選題：Supportvectormachinescanbeusedforclassificationandregression.

選項：

A、正確

B、錯誤

答案：【正確】10.單選題：Outlierminingsuchasdensitybasedmethodbelongstosupervisedlearning.

選項：

A、正確

B、錯誤

答案：【錯誤】CourseworkAnalysisofDrivingBehavior1.Inthiscoursework,youarerequiredtousetechniquesofdataminingtostudytheabnormaldrivingbehavior.Pleasedownloadtheattachmentandreadthedetailinformationofthecourseworkincoursework.docxfile.Youneedtochooseonetodofromtask1andtask2,andthenchooseonetodofromtask3andtask4.Hopeyougetgoodunderstandingafterlearningthiscourse.

Completestructureandclearcitationstandard.CleanvaluelessdatainVBOX.csv,oversamplingorundersamplingthedataifsupervisedlearningtaskwaschoosed.Fortask1ortask2,suitablemethods,clearsteps,andgoodpredictionresults.Fortask3ortask4,thesolutionproposedshoulebereasonableandfeasible.Week2.Datapre-processingTest21.單選題：Whichoneiswrongwaytonormalizedata?

選項：

A、Min-maxnormalization

B、Simplescaling

C、Z-scorenormalization

D、Normalizationbydecimalscaling

答案：【Simplescaling】2.單選題：WhichoneiswrongaboutEqual-width(distance)partitioningandEqual-depth(frequency)partitioning?

選項：

A、Equal-widthpartitioningisthemoststraightforward,butoutliersmaydominatepresentation.

B、Equal-depthpartitioningdividestherangeintoNintervals,eachcontainingapproximatelysamenumberofsamples.

C、Theintervaloftheformeroneisnotequal.

D、Thenumberoftuplesisthesamewhenusingthelatterone.

答案：【Theintervaloftheformeroneisnotequal.】3.單選題：Whichoneiswrongaboutmethodsfordiscretization?

選項：

A、HistogramanalysisandBingingarebothunsupervisedmethods.

B、Clusteringanalysisonlybelongstotop-downsplit.

C、Intervalmergingbyc2Analysiscanbeappliedrecursively.

D、Decision-treeanalysisisEntropy-baseddiscretization.

答案：【Clusteringanalysisonlybelongstotop-downsplit.】4.單選題：HowtoconstructnewfeaturespacebyPCA?

選項：

A、NewfeaturespacebyPCAisconstructedbychoosingthemostimportantfeaturesyouthink.

B、NewfeaturespacebyPCAisconstructedbynormalizinginputdata.

C、NewfeaturespacebyPCAisconstructedbyselectingfeaturesrandomly.

D、NewfeaturespacebyPCAisconstructedbyeliminatingtheweakcomponentstoreducethesizeofthedata.

答案：【NewfeaturespacebyPCAisconstructedbyeliminatingtheweakcomponentstoreducethesizeofthedata.】5.單選題：Whichisnotthemajortasksindatapreprocessing?

選項：

A、Clean

B、Integration

C、Transition

D、Reduction

答案：【Transition】6.單選題：Whichisnotthereasonweneedtopreprocessthedata?

選項：

A、tosavetime

B、tomakeresultmeetourhypothesis

C、toavoidunreliableoutput

D、toeliminatenoise

答案：【tomakeresultmeetourhypothesis】7.多選題：Whicharethecommonusedwaystosampling?

選項：

A、Simplerandomsamplewithoutreplacement

B、Simplerandomsamplewithreplacement

C、Stratifiedsample

D、Clustersample

答案：【Simplerandomsamplewithoutreplacement;Simplerandomsamplewithreplacement;Stratifiedsample;Clustersample】8.多選題：Whichoneisrightaboutwavelettransforms?

選項：

A、Wavelettransformsstorelargefractionsofthestrongestofthewaveletcoefficients.

B、TheDWTdecomposeseachsegmentoftimeseriesviathesuccessiveuseoflow-passandhigh-passfilteringatappropriatelevels.

C、Wavelettransformscanbeusedforreducingdataandsmoothingdata.

D、Wavelettransformsmeansapplyingtopairsofdata,resultingintwosetofdataofthesamelength.

答案：【TheDWTdecomposeseachsegmentoftimeseriesviathesuccessiveuseoflow-passandhigh-passfilteringatappropriatelevels.;Wavelettransformscanbeusedforreducingdataandsmoothingdata.】9.多選題：Whicharetherightwaytohandlenoisedata?

選項：

A、Regression

B、Cluster

C、WT

D、Manual

答案：【Regression;Cluster;WT;Manual】10.多選題：Whicharetherightwaytofillinmissingvalues?

選項：

A、Smartmean

B、Probablevalue

C、Ignore

D、Falsify

答案：【Smartmean;Probablevalue;Ignore】11.單選題：Discretizationmeansdividingtherangeofacontinuousattributeintointervals.

選項：

A、正確

B、錯誤

答案：【正確】Week3.InstancebasedlearningTest31.單選題：What'sthedifferencebetweeneagerlearnerandlazylearner?

選項：

A、Eagerlearnerswouldgenerateamodelforclassificationwhilelazylearnerwouldnot.

B、Eagerlearnersclassifytheturplebasedonitssimilaritytothestoredtrainingturplewhilelazylearnernot.

C、Eagerlearnerssimplystoredata(ordoesonlyalittleminorprocessing)whilelazylearnernot.

D、Lazylearnerwouldgenerateamodelforclassificationwhileeagerlearnerwouldnot.

答案：【Eagerlearnerswouldgenerateamodelforclassificationwhilelazylearnerwouldnot.】2.多選題：What’sthemajorcomponentsinKNN?

選項：

A、Howtomeasuresimilarity?

B、Howtochoose"k"?

C、Howareclasslabelsassigned?

D、Howtodecidethedistance?

答案：【Howtomeasuresimilarity?;Howtochoose"k"?;Howareclasslabelsassigned?】3.多選題：HowtochoosetheoptimalvalueforK?

選項：

A、Cross-validationcanbeusedtodetermineagoodvaluebyusinganindependentdatasettovalidatetheKvalues.

B、LowvaluesforK(likek=1ork=2)canbenoisyandsubjecttotheeffectofoutliers.

C、Alargekvaluecanreducetheoverallnoisesothevaluefor'k'canbeasbigaspossible.

D、Historically,theoptimalKformostdatasetshasbeenbetween3-10.

答案：【Cross-validationcanbeusedtodetermineagoodvaluebyusinganindependentdatasettovalidatetheKvalues.;LowvaluesforK(likek=1ork=2)canbenoisyandsubjecttotheeffectofoutliers.;Historically,theoptimalKformostdatasetshasbeenbetween3-10.】4.多選題：WhichoneofthefollowingwayscanbeusedtoobtainattributeweightforAttribute-WeightedKNN?

選項：

A、Priorknowledge/experience.

B、PCA,FA(Factoranalysismethod).

C、Informationgain.

D、Gradientdescent,simplexmethodsandgeneticalgorithm.

答案：【Priorknowledge/experience.;PCA,FA(Factoranalysismethod).;Informationgain.;Gradientdescent,simplexmethodsandgeneticalgorithm.】5.單選題：Thewaytoobtaintheclassificationforanewinstancefromtheknearestneighborsistocalculatethemajorityclassofkneighbors.

選項：

A、正確

B、錯誤

答案：【正確】6.單選題：Thewaytoobtaintheregressionforanewinstancefromtheknearestneighborsistocalculatetheaveragevalueofkneighbors.

選項：

A、正確

B、錯誤

答案：【正確】7.單選題：DatanormalizationbeforeMeasureDistancecanavoiderrorscausedbydifferentdimensions,self-variations,orlargenumericaldifferences.

選項：

A、正確

B、錯誤

答案：【正確】8.單選題：ByEuclideandistanceorManhattandistance,wecancalculatethedistancebetweentwoinstances.

選項：

A、正確

B、錯誤

答案：【正確】9.單選題：Normalizingthedatacansolvetheproblemthatdifferentattributeshavedifferentvalueranges.

選項：

A、正確

B、錯誤

答案：【正確】10.單選題：AtclassificationstageKNNwouldstoreallinstanceorsometypicalofthem.

選項：

A、正確

B、錯誤

答案：【錯誤】11.單選題：AtlearningstageKNNwouldfindtheKclosestneighborsandthendecideclassifyKidentifiednearestlabel.

選項：

A、正確

B、錯誤

答案：【錯誤】12.單選題：ThewaytoobtaininstanceweightforDistance-WeightedKNNistocalculatethereciprocalofthedistancesquaredbetweenobjectandneighbors.

選項：

A、正確

B、錯誤

答案：【正確】Week4.DecisionTreesTest41.多選題：Whichdescriptionisrightaboutnodesindecisiontree?

選項：

A、Internalnodestestthevalueofparticularfeatures

B、Leafnodesspecifytheclass

C、Branchnodesdecidetheresult

D、Rootnodesdecidethestartpoint

答案：【Internalnodestestthevalueofparticularfeatures;Leafnodesspecifytheclass】2.多選題：Post-pruninginCARTconsistsofthefollowingprocedure:

選項：

A、First,considerthecostcomplexityofatree.

B、Then,foreachinternalnode,N,computethecostcomplexityofthesubtreeatN.

C、AndalsocomputethecostcomplexityofthesubtreeatNifitweretobepruned.

D、Atlast,comparethetwovalues.IfpruningthesubtreeatnodeNwouldresultinasmallercostcomplexity,thesubtreeispruned.Otherwise,thesubtreeiskept.

答案：【First,considerthecostcomplexityofatree.;Then,foreachinternalnode,N,computethecostcomplexityofthesubtreeatN.;AndalsocomputethecostcomplexityofthesubtreeatNifitweretobepruned.;Atlast,comparethetwovalues.IfpruningthesubtreeatnodeNwouldresultinasmallercostcomplexity,thesubtreeispruned.Otherwise,thesubtreeiskept.】3.多選題：Whichoneisrightaboutpre-pruningandpost-pruning?

選項：

A、Bothofthemaremethodstodealwithoverfittingproblem.

B、Pre-pruningdoesnotsplitanodeifthiswouldresultinthegoodnessmeasurefallingbelowathreshold.

C、Post-pruningremovesbranchesfroma“fullygrown”tree.

D、Thereisnoneedtochooseanappropriatethresholdwhenmakingpre-pruning.

答案：【Bothofthemaremethodstodealwithoverfittingproblem.;Pre-pruningdoesnotsplitanodeifthiswouldresultinthegoodnessmeasurefallingbelowathreshold.;Post-pruningremovesbranchesfroma“fullygrown”tree.】4.多選題：Whichoneisrightaboutunderfittingandoverfitting?

選項：

A、Underfittingmeanspooraccuracybothfortrainingdataandunseensamples.

B、Overfittingmeanshighaccuracyfortrainingdatabutpooraccuracyforunseensamples.

C、Underfittingimpliesthemodelistoosimplethatweneedtoincreasethemodelcomplexity.

D、Overfittingoccurstoomanybranchesthatweneedtodecreasethemodelcomplexity.

答案：【Underfittingmeanspooraccuracybothfortrainingdataandunseensamples.;Overfittingmeanshighaccuracyfortrainingdatabutpooraccuracyforunseensamples.;Underfittingimpliesthemodelistoosimplethatweneedtoincreasethemodelcomplexity.;Overfittingoccurstoomanybranchesthatweneedtodecreasethemodelcomplexity.】5.多選題：Whichisthetypicalalgorithmstogeneratetrees?

選項：

A、ID3

B、C4.5

C、CART

D、PCA

答案：【ID3;C4.5;CART】6.多選題：ComputinginformationgainforcontinuousvalueattributewhenusingID3consistsofthefollowingprocedure:

選項：

A、SortthevalueAinincreasingorder.

B、Considerthemidpointbetweeneachpairofadjacentvaluesasapossiblesplitpoint.

C、Selecttheminimumexpectedinformationrequirementasthesplit-point.

D、Split.

答案：【SortthevalueAinincreasingorder.;Considerthemidpointbetweeneachpairofadjacentvaluesasapossiblesplitpoint.;Selecttheminimumexpectedinformationrequirementasthesplit-point.;Split.】7.單選題：ID3useinformationgainasitsattributeselectionmeasure.AndtheattributewiththelowestinformationgainischosenasthesplittingattributefornoteN.

選項：

A、正確

B、錯誤

答案：【錯誤】8.單選題：Ruleiscreatedforeachpartfromitsroottoitsleafnotes.

選項：

A、正確

B、錯誤

答案：【正確】9.單選題：GainratioisusedasattributeselectionmeasureinC4.5andtheformulaisGainRatio(A)=Gain(A)/SplitInfo(A).

選項：

A、正確

B、錯誤

答案：【正確】10.單選題：ThecostcomplexitypruningalgorithmusedinCARTevaluatecostcomplexitybythenumberofleavesinthetree,andtheerrorrate.

選項：

A、正確

B、錯誤

答案：【正確】Week5.SupportVectorMachineTest51.多選題：WhichoneisrightabouttheadvantagesofSVM?

選項：

A、Theyareaccurateinhigh-dimensionalspaces.

B、Theyarememoryefficient.

C、Thealgorithmisnotproneforover-fittingcomparedtootherclassificationmethod.

D、Thesupportvectorsaretheessentialorcriticaltrainingtuples.

答案：【Theyareaccurateinhigh-dimensionalspaces.;Theyarememoryefficient.;Thealgorithmisnotproneforover-fittingcomparedtootherclassificationmethod.;Thesupportvectorsaretheessentialorcriticaltrainingtuples.】2.多選題：What'stheproblemofOVR?

選項：

A、Sensitivetotheaccuracyoftheconfidencefiguresproducedbytheclassifiers.

B、Thescaleoftheconfidencevaluesmaydifferbetweenthebinaryclassifiers.

C、Thebinaryclassificationlearnersseeunbalanceddistributions.

D、Onlywhentheclassdistributionisbalancedcanbalanceddistributionsattain.

答案：【Sensitivetotheaccuracyoftheconfidencefiguresproducedbytheclassifiers.;Thescaleoftheconfidencevaluesmaydifferbetweenthebinaryclassifiers.;Thebinaryclassificationlearnersseeunbalanceddistributions.】3.多選題：WhatadaptationscanbemadetoallowSVMtodealwithMulticlassClassificationproblem?

選項：

A、Oneversusrest(OVR).

B、Oneversusone(OVO).

C、Errorcorrectinginputcodes(ECIC).

D、Errorcorrectingoutputcodes(ECOC).

答案：【Oneversusrest(OVR).;Oneversusone(OVO).;Errorcorrectingoutputcodes(ECOC).】4.多選題：Whichisthetypicalcommonkernel?

選項：

A、Linear

B、Polynomial

C、Radialbasisfunction(Gaussiankernel)

D、Sigmoidkernel

答案：【Linear;Polynomial;Radialbasisfunction(Gaussiankernel);Sigmoidkernel】5.多選題：WhatthefeatureofSVM?

選項：

A、Extremelyslow,butarehighlyaccurate.

B、Muchlesspronetooverfittingthanothermethods.

C、Blackboxmodel.

D、Provideacompactdescriptionofthelearnedmodel.

答案：【Extremelyslow,butarehighlyaccurate.;Muchlesspronetooverfittingthanothermethods.;Provideacompactdescriptionofthelearnedmodel.】6.單選題：Ifyouhaveabigdataset,SVMissuitableforefficientcomputation.

選項：

A、正確

B、錯誤

答案：【錯誤】7.單選題：Regressionformulasincludingthreetypes:linear,nonlinearandgeneralform.

選項：

A、正確

B、錯誤

答案：【正確】8.單選題：Errorcorrectingoutputcodes(ECOC)isakindofproblemtransformationtechniques.

選項：

A、正確

B、錯誤

答案：【錯誤】9.單選題：ThereisnostructuredwayandnogoldenrulesforsettingtheparametersinSVM.

選項：

A、正確

B、錯誤

答案：【正確】10.單選題：Kerneltrickwasusedtoavoidcostlycomputationanddealwithmappingproblems.

選項：

A、正確

B、錯誤

答案：【正確】Week6.OutlierMiningTest61.多選題：Whichoneisrightaboutthreemethodsofoutliermining?

選項：

A、Statistics-basedapproachissimpleandfastbutdifficulttodealwithperiodicitydataandcategoricaldata.

B、Theefficiencyofdistance-basedapproachislowforthegreatdatasetinhighdimensionalspace.

C、Distance-basedapproachcannotbeusedinmultidimensionaldataset.

D、Density-basedapproachspendslowcostonsearchingneighborhood.

答案：【Statistics-basedapproachissimpleandfastbutdifficulttodealwithperiodicitydataandcategoricaldata.;Theefficiencyofdistance-basedapproachislowforthegreatdatasetinhighdimensionalspace.】2.多選題：Howtopicktherightkbyaheuristicmethodfordensity-basedoutlierminingmethod?

選項：

A、Kshouldbeatleast10toremoveunwantedstatisticalfluctuations.

B、Pick10to20appearstoworkwellingeneral.

C、Picktheupperboundvalueforkasthemaximumof“closeby”objectsthatcanpotentiallybeglobaloutliers.

D、Picktheupperboundvalueforkasthemaximumof“closeby”objectsthatcanpotentiallybelocaloutliers.

答案：【Kshouldbeatleast10toremoveunwantedstatisticalfluctuations.;Pick10to20appearstoworkwellingeneral.;Picktheupperboundvalueforkasthemaximumof“closeby”objectsthatcanpotentiallybelocaloutliers.】3.多選題：Whichoneisthemethodtodetectoutliers?

選項：

A、Statistics-basedapproach

B、Distance-basedapproach

C、Bulk-basedapproach

D、Density-basedapproach

答案：【Statistics-basedapproach;Distance-basedapproach;Density-basedapproach】4.多選題：Whatisapplicationcaseofoutliermining?

選項：

A、Trafficincidentdetection

B、Creditcardfrauddetection

C、Networkintrusiondetection

D、Medicalanalysis

答案：【Trafficincidentdetection;Creditcardfrauddetection;Networkintrusiondetection;Medicalanalysis】5.多選題：Whichdescriptionisrighttodescribeoutliers?

選項：

A、Outlierscausedbymeasurementerror

B、Outliersreflectinggroundtruth

C、Outlierscausedbyequipmentfailure

D、Outliersneededtobedroppedoutalways

答案：【Outlierscausedbymeasurementerror;Outliersreflectinggroundtruth;Outlierscausedbyequipmentfailure】6.單選題：Distance-basedoutlierMiningisnotsuitabletodatasetthatdoesnotfitanystandarddistributionmodel.

選項：

A、正確

B、錯誤

答案：【錯誤】7.單選題：Anoutlierisadataobjectthatdeviatessignificantlyfromtherestoftheobjects,asifitweregeneratedbyadifferentmechanism.

選項：

A、正確

B、錯誤

答案：【正確】8.單選題：MahalanobisDistanceaccountsfortherelativedispersionsandinherentcorrelationsamongvectorelements,whichisdifferentfromEuclideanDistance.

選項：

A、正確

B、錯誤

答案：【正確】9.單選題：Whenidentifyingoutlierswithadiscordancytest,thedatapointisconsideredasanoutlierifitfallswithintheconfidenceinterval.

選項：

A、正確

B、錯誤

答案：【錯誤】10.單選題：Statistic-basedmethodneedstorequireknowingthedistributionofthedataandthedistributionparametersinadvance.

選項：

A、正確

B、錯誤

答案：【正確】Week7.EnsembleLeaningTest71.多選題：Whichstepisnecessarywhenconstructinganensemblemodel?

選項：

A、Creatingmultipledataset

B、Constructingasetofclassifiersfromthetrainingdata

C、Combiningpredictionsmadebymultipleclassifierstoobtainfinalclasslabel

D、Findthebestperformingpredictionstoobtainfinalclasslabel

答案：【Creatingmultipledataset;Constructingasetofclassifiersfromthetrainingdata;Combiningpredictionsmadebymultipleclassifierstoobtainfinalclasslabel】2.多選題：Whichoneisrightwhendealingwiththeclass-imbalanceproblem？

選項：

A、Oversamplingworksbydecreasingthenumberofminoritypositivetuples.

B、Undersamplingworksbyincreasingthenumberofmajoritynegativetuples.

C、Smotealgorithmaddssynthetictuplesthatareclosetotheminoritytuplesintuplespace.

D、Threshold-movingandensemblemethodswereempiricallyobservedtooutperformoversamplingandundersampling.

答案：【Smotealgorithmaddssynthetictuplesthatareclosetotheminoritytuplesintuplespace.;Threshold-movingandensemblemethodswereempiricallyobservedtooutperformoversamplingandundersampling.】3.多選題：Howtodealwithimbalanceddatain2-classclassification?

選項：

A、Oversampling

B、Undersampling

C、Threshold-moving

D、Ensembletechniques