eventse52se.com

[Typetext][Typetext][Typetext]2014TradeScienceInc.
ISSN:0974-7435Volume10Issue20BioTechnologyAnIndianJournalFULLPAPERBTAIJ,10(20),2014[12660-12666]AcomparativestudyoncarevaluationforecastbasedondataminingYuqingYuan1*,LijunLing2,XianglingKuang1,QinggangZuo31SchoolofEconomicsandManagement,HubeiUniversityofAutomotiveTechnology,ShiYan,442002,(CHINA)2InstituteofScienceandTechnology,HubeiUniversityofAutomotiveTechnology,ShiYan,442002,(CHINA)3DepartmentofScienceandTechnology,ShiyanCentralSub-branchofthePeople'sBankofChinaShiya,People'sRepublicofChina,ShiYan,442002,(CHINA)E-mail:yyq115805@163.
comABSTRACTLogisticregression(LR),artificialneuralnetwork(ANN),decisiontrees(DT),andsupportvectormachine(SVM)wereusedinforecastingcaracceptability,andtheiraccuracy,sensitivityandspecificitywerecompared.
Theresultsshowthatsupportvectormachine(SVM)modelcanwellpredictthecaracceptabilityevaluationwith99.
62percentofaccuracyrateand100percentofsensitivityandspecificity.
Thefactorofsecurityhasthemostinfluenceoncaracceptabilityevaluation.
Comparativestudymethodissuitablefortheevaluationofcaracceptabilityforecasting,canalsobeextendedtoallotherareas.
KEYWORDSDatamining;Caracceptability;Logisticregression;Supportvectormachine;Artificialneuralnetworks;Decisiontree.
BTAIJ,10(20)2014YuqingYuanetal.
12661INTRODUCTIONTheautomotiveindustryisthepillarindustryofthenationaleconomy,andmillionsofpeoplearecloselyrelatedtoit.
Formostpeople,buyingacaristobuyahouseoutsideinadditiontoamaximumconsumption,canbetterreflectconsumerdemandandtherealmarketbehaviour.
Whenconsumersconsiderwhenbuyingacar,therearemanyfactorsthatcouldrestricttheirchoice,suchasprice,carperformance,thecar'scomfortandsafety,etc.
Thesefactorsformtherightcarevaluation.
Thefiercemarketcompetitionforcedautocompaniesinaveryshortdevelopmentcycleofcontinuousimprovementandinnovationinproductdesigntomeettheneedsofhighlydiversetargetmarket[4-6].
Therefore,areasonableevaluationmethodisequallyimportantforcarconsumersandproducers.
Itcannotonlyreducetheburdenondealers,butalsoincreasesales.
Inaddition,itplaysastrategicrole,canimprovecustomerservicelevelsinahighlycompetitivemarketenvironment[1-3]ByunproposedextensionofAHPselectvehiclepurchasepatterns.
Laiandsomehaveproposedamethodtohelpdesignersimprovethequalityfeelofautomotiveproducts.
AlnoukariandAlhussanproposedusingdataminingtechniquestopredictthefutureoftheautomotivemarketdemand.
Chenandsomepeoplemakeuseofartificialintelligencemethods;thepracticalproblemsfacedbytheautopartsindustryinproductperformanceobjectivelyclassify[19-21].
Dataminingtechniquesfromthehiddendatafoundusefulinformationtofacilitatesmartsummaryandfuturedecisions,soithasgreatvisibilityinareasofresearchandcommercialareas,invariousapplications,includingmanufacturing,marketing,finance,healthcareandotherfieldshaveoutstandingperformance,isadataconversionindispensabletoolforinformation.
Advantagesofdataminingtechnologyisitsabilitytohandlemassivetrafficdatainordertoadapttomarketchanges,canprovidedecisionmakerswithapowerfultool.
Itiswidelyusedinbusinessmanagement,governmentadministration,scientificandengineeringdatamanagementoflargeamountsofdataprocessing.
Withtheexplosivegrowthofdata,dataminingtechniquesandtoolshavebecomeanurgentneed,itwillbeprocesseddataintelligentlyandautomaticallyconvertedintousefulinformationandknowledge.
Itimprovestheircompetitiveadvantageandincreasesthecompany'srevenue,butalsoenablesenterprisestoprovidebetterservicetoretaincustomers.
Inthepastfewyears,theclassicdataminingtechnologysuchaslogisticregression(LR),artificialneuralnetworks(ANN),decisiontrees(DT)andSupportVectorMachine(SVM)havebeensuccessfullyappliedinmanyfieldstosolvepracticalproblemsofproduction,salesandresearchinemerging.
[18]However,nocomparativestudytoassessaproduct'sacceptabilityforprediction.
Accurateassessmentofthedevelopmentofproductacceptabilityhasbecomeanimportantresearchtopic.
Thepurposeofthisstudyistoprovideamethodforcomparativeevaluationofthestudytoassessthepredictiveevaluationoftheautomotive,andthenextendedtootherfields.
Bymodellingrespectivelyusinglogisticregression(LR),artificialneuralnetwork(ANN),decisiontrees(DT),supportvectormachine(SVM)thesetechnologyforecastingautomotive,andtheiraccuracy,sensitivityandspecificitywerecompared.
Theresultsshowthatsupportvectormachine(SVM)technologyhasmadethebestassessmentandprediction.
LargelyduetotheperformanceofSVMdependsonthechoiceofkernelfunction,thepaperfinallylinearkernel,polynomialkernel,radialbasis(RBF)kernelandtheS-shapedcorekernelcomparativestudy,polynomialkernelhasachievedthebestresults.
THEBASICCONCEPTANDRESEARCHFRAMEWORKLogisticregressionLogisticregressionisapopularnon-linearstatisticalmodel,andiswidelyusedinmanyfields.
Comparedwiththemultipleregressionmodels,Logisticregressionmodelcansimulatetwoormoredependentvariables.
Forbinaryvariablescanbedefinedasaneventofinterestcodingandcodingarenotinterestedintheeventof0[7-11]Alogisticregressionmodelcanbewrittenas:1122(1)log()1(1)kkpYXXXpYEquationcanbemodifiedasfollows:1(1)1zpYeWhere1122kkzXXXThelogisticregressionmodelenablesustocalculatetheprobabilityofeventY=1occurringforeachcase.
Thepredictors,Xkcanbeamixtureofcontinuousandcategoricalvariables.
12662AcomparativestudyoncarevaluationforecastbasedondataminingBTAIJ,10(20)2014DecisiontreeAdecisiontreeisapredictivemodel;Itrepresentsamappingbetweenobjectattributesandobjectvalues.
Eachnodeinthetreerepresentsanobject,andapossibleattributevaluesforeachforkedpathsarerepresented,eachleafnodecorrespondstothevalueofthepathfromtherootnodetotheleafnodeexperiencedrepresentedobject.
Treeonlyasingleoutput,ifitwanttohaveapluralityofoutput,itcanestablishanindependentdecisiontreetohandledifferentoutput.
DataMiningDecisionTreeisatechniqueoftenused,canbeusedtoanalyzethedata,alsocanbeusedtomakepredictions[12-13].
CommontreealgorithmCHAID(Chi-squaredAutomaticcross-checking),CART(ClassificationandRegressionTrees)andC5.
0.
CARTalgorithmusestheGiniasastandarddecisiontreesplit,C5.
0entropyasthesplitcriteria,CHAIDusingchi-squaretestassegmentationcriteria.
Throughthesealgorithmwillgeneratingtreediagram,,splittingruleandimportantinformationcanbereflectedfromtheFigureout.
ArtificialneuralnetworksArtificialneuralnetworkisanapplicationsimilartothestructureofthebrain-Fimathematicalmodelofinformationprocessing,constitutedbyalargenumberofinterconnectednodes(orneurons)between.
Eachnoderepresentsaspecificoutputfunction,calledactivationfunction.
Eachconnectionbetweentwonodesrepresentsaweightedvalueoftheconnectionforthesignal,referredtotheweight,itisequivalenttothememoryofartificialneuralnetworks.
Theoutputofthenetworkaccordingtothedifferentnetworkconnections,weightsandexcitationfunctionsanddifferent.
MultilayerPerception(MLP)isthemostwidelyusedneuralnetworkmodelinthedataanalysis,itwillentermultipledatasetsaremappedtoasingleoutputdataset.
Artificialneuralnetworkcanidentifyandstudythepatternofassociationbetweeninputdatasetandthecorrespondingtargetvalue[14-15].
However,artificialneuralnetworks(ANNs)forits"blackbox"approachandinterpretationdifficultiessuffercriticism.
Nevertheless,comparedwithothercomparativeclassificationtechniques,artificialneuralnetworktoprovidealternativemodels.
Aftertraining,theartificialneuralnetworkcanbeusedtopredictindependentinputdataofthenew.
SupportvectormachineSupportvectormachine(SVM)isasupervisedlearningmethodcanbewidelyusedinstatisticalclassificationandregressionanalysis.
Supportvectormachineisaclassclassifierwithdifferentkindofsamplescanbeseparatedinthesamplespacehyperplane.
Thatisagivennumberofmarkedwelltrainingsamples.
SVMalgorithmoutputsanoptimizedseparatinghyperplane.
TheessenceofSVMalgorithmistofindacanbeavaluemaximizinghyperplane,thisvalueistheminimumdistancehyperplanedistanceofallthetrainingsamples.
Theminimumdistanceiscalledinterval(margin)[16-17].
Thefollowingequationdefinesahyperplaneexpression:0()TfxxWhichiscalledtheweightvector0calledbias.
Whereinxrepresentsfromthosepointsclosesthyperplane,thesepointsarecalledsupportvectors.
ThekeytoSVMisthekernelfunction.
Vectorsetoflow-dimensionalspaceisoftendifficultdivision,sothesolutionistomapthemtothehigh-dimensionalspace.
Butthedifficultyofthisapproachistobringthecomputationalcomplexityincreases,whilekerneljustingenioussolvethisproblem[15].
Inotherword,aslongastheselectionofappropriatekernelfunction,youcangetahigh-dimensionalspaceclassificationfunction.
InSVMtheory,usingdifferentkernelfunctionwillleadtoadifferentSVMalgorithmtogetadifferentoutput.
ResearchframeworkComparativestudiesofdifferentalgorithmstoassesspredictivecapabilities,thisstudyprovideaverygoodsolutiontothispracticalproblemcaracceptabilityevaluation.
ResearchframeworkshowninFigure1,eachstageoftheprocessisasfollows:Figure1:ResearchFrameworkBTAIJ,10(20)2014YuqingYuanetal.
12663Collectionandinputrawdata:Itcomprisesacollectionoftheoriginaldataandselectsacaracceptabilityevaluationcharacteristicparameter.
Datapre-processing:First,thedatausedtocalculatethenominaldataformat.
Secondly,thecarassesstheacceptabilityofthedatasetisdividedintofourcategories(unacceptable,acceptable,goodandverygood),inthepresentstudy,inordertosimplifythecomplexityoftheacceptabilityoftheresearchintotwocategories,acceptableandunacceptable,thegoodkindofthesamenatureandverygoodclassmergingtoacceptableclass.
ModellingResearch:Studiesusinglogisticregression(LR),artificialneuralnetwork(ANN)anddecisiontree(DT),supportvectormachine(SVM)offourkindsofalgorithmtocalculatetheevaluationofforecastingaccuracyrate,sensitivityandspecificity.
Accuracy,sensitivityandspecificityofthetestmethodisasfollows:Process1collectionandinputrawdataset:Itincludesthecollectionofrawdata,selectingthedataandfocusingonthefeaturesinfluencethecarevaluation.
Process2pre-processingthedataset:Thisstepincludesthreeparts.
Firstly,thedataaretransferredtoforms"nominaltonumeric"forcalculating.
Secondly,therearefourclasses(unacceptable,acceptable,good,andvery-good)incarevaluationdataset.
Inthisstudy,wecombinedthesimilarclasses(acceptable,goodandvery-good)intooneclass.
Thefourclasseswerecombinedtoformtwoclasses(unacceptable,acceptable).
Process3modellingtraining:Studiesusinglogisticregression(LR),artificialneuralnetwork(ANN),decisiontree(DT)andsupportvectormachine(SVM)intotalfourkindsofalgorithmtocalculatetheevaluationofforecastingaccuracyrate,sensitivityandspecificity.
Accuracy,sensitivityandspecificityofthetestmethodisasfollows:TPTNAccuracyTPTNFPFNTPSensitivityTPFNTNSpecificityTNFPIfaninstanceoftheclassispositiveandalsopredictedpositiveclass,thatisthetruepositive,iftheinstanceoftheclassispredictednegativepositiveclass,calledfalsepositive.
Accordingly,iftheinstanceisnegativeclassispredictedtobecomethenegativeclass,calledtruenegative,thepositiveclassispredictedtobecomethenegativeclassisfalsenegative.
THEEMPIRICALRESEARCHDatadescriptionInthiswork,areal-worldcarevaluationdatabasewastakenfromtheUCIrepositoryofmachinelearningdatabaseasdescribedinTABLE1.
Itcontains1728instancesandclassifiedintofourclasses,thereisnomissingvalueinthedataset.
Thecarevaluationdatabasecontainssixattributesexampleswithacar(Buying,Main,Doors,Persons,LugbootandSafety)TABLE1:CarattributeDescriptionClementinemodellingThepentagon-shapednodesshowtheconstructionofthemodelsusinglogisticregression,decisiontrees(CART)andneuralnetwork.
Thediamond-shapednodesshowthemodeloutputsoftherespectivemodels.
Forthelogisticregressionmodel,fourselectionmethods(ENTER,STEPWISE,FORWARDS,BACKWARDS)werecomparedusingtheAnalysisandEvaluationnodes.
Whilefordecisiontress,theC5.
0,CHAIDandCARTmodelsweregeneratedandcompared.
Then,thethreepredictivemodelswhicharestepwiselogisticregression,CARTandneuralnetworkareconnectedtothe"analysis"nodewhichprovidesthecomputationofaccuracyrates,whiletheevaluationnodeproducestheliftcharts.
12664AcomparativestudyoncarevaluationforecastbasedondataminingBTAIJ,10(20)2014Figure2:DataminingprocessflowdiagramComparisonofresultsThedifferentmodellingalgorithmresultsasshownintheTABLE2:TABLE2:ComparisonofmodellingresultsFourdifferentlogisticregressionmethodshavethesameaccuracy,sensitivityandspecificity.
Italsoshowsthatthelogisticregressionmodeltoassessthecaracceptabilityisnotsignificant.
Decisiontreeisthemosteasilyunderstoodmodel,andcanbeeasilyconvertedintoasetofrules.
Inaddition,decisiontreealgorithmcanhandlebothdiscreteandcontinuousdata,withouttheneedfordatatomakeaprioriassumptions.
Becauseoftheseadvantages,themethodofdecisiontreeiswidelyusedforclassificationandprediction.
Thetableshowsthedifferencebetweenthreekindsofdecisiontreemodelaccuracy,sensitivityandspecificity.
Sensitivityconsideredtruepositiverate,referstotheactualacceptabilityoftheprobabilityisdeterminedtobeaccepted.
Andspecificityisconsideredthetruepositiverate,referstotheactualacceptabilityofnottoaccepttheprobabilityisdeterminedunacceptable.
Theresultsofthreedecisiontreealgorithmsareveryclose,butCARTmodelhasthebesttestingandprediction.
Artificialneuralnetworkscanbeusedtopredictcomplexsystemsisdeterminedbytherelationshipbetweenthenumberoftrainingsamples,thetrainingsamplesandthetestsamples,meanwhile,theforecastperformancealsodependsonthechoiceofdatastructures,dataqualityandvariables.
Inthepresentstudy,artificialneuralnetworkachievedgoodperformance,secondonlytoSVM.
Inallthesemodels,SVMhasbetterpredictiveability,whichshowedthebestaccuracy,sensitivityandspecificity.
TheSVMpredictionresultsaredirectlyrelatedtothechoiceofkernel,SecondarymodellingfourdifferentkernelfunctionsofSVMshowninFigure3,experimentalresultsshowthatthepolynomialkernelfunctionhasthebestpredictiveability.
Figure3:SVMkernelfunctionmodelingflowInsummary,thisstudySVMpolynomialkernelfunctiontopredictthecaracceptability,withaverygoodperformance,trainingandtestsetsofaccuracywere99.
92%,99.
62%,sensitivityis100%,specificityof99.
88%and100%asshowninTABLE2.
BTAIJ,10(20)2014YuqingYuanetal.
12665TABLE3:PerformanceofthedifferentkernelsTheimportanceofthevariablesanalyzed,fourmodelresultswerealmostidentical,carsecurityparametersareconsideredthemostsignificantimpactontheacceptabilityofthecarthemost,followedbythecarcanaccommodatethenumber.
Thetwoparameteronconsumeracceptabilityoftheeffectsarelesssignificantisthenumberandsizeofthetrunkdoor.
TABLE4:ComparisonoftheimportanceofautovariablesConclusionFourmodelsusedintheempiricalcaracceptabilityevaluationwerecomparedandresearched,theresultsshowedthatallfourmodelshavesimilargoodpredictiveability,andsupportvectormachinemodel(polynomialkernel)showedthebestaccuracy,sensitivityandspecificity,withthebestpredictiveability,canbeverygoodforcaracceptabilityevaluation.
Inthesixattributesofthecar,thesafetyhasthelargestinfluenceoncaracceptabilityfollowedbyoccupancy,however,consumersislesssensitivetothesetwofactorsofthesizeofthetrunkandthenumberofdoor.
Thiscanhelpcompaniesmakebetterpolicy,targeted,improvedmethodstoimprovethecar'sacceptabilityandconsumersatisfaction.
Summinguptheappeal,theuseofdataminingmodellingmethodcanaccuratelypredicttheacceptabilityofthecarinordertobuildagoodbridgebetweenconsumersandbusinesses,fortheenterpriseprofitsandconsumers'satisfaction.
CONCLUSIONSAcceptanceoftheproductofgrowingconcern,themanufacturermustknowwhichfactorsinfluenceconsumers'buyingdecisions.
Inrecentyears,theproducthasbeenin-depthevaluationoftheacceptabilityofresearch,unfortunately,manufacturersoftenmisunderstandtherealneedsofconsumers,andhowtobetterevaluatetheacceptabilityoftheproductisthekeyissueofproductdevelopment.
Inthisthesis,fortheevaluationofvehicleslinesempiricalresearch,usingfourmodelsfortheevaluationofautomotiveforecastingacomparativestudy,theexperimentalresultsshowthat,usingpolynomialkernelSVMmodelcanbestbeassessedoncarevaluationprediction.
Intheevaluationofthecar,thesafetyperformanceofthemostsignificantfactors,occupancysecond,butcustomerhavenospecialrequirementsofthesizeandthenumberofdoor.
Allinall,acomparativestudydifferentfourmodelsbadeondataminingtoevaluatecaracceptability,resultsshowed,SVMmodelcanbettersolvethecarevaluation,inturn,andthemethodcanbeextendedtootherindustriestosolvetheevaluationoftheproduct.
REFERENCES[1]LiboLi,FrankGoethals,AntonioGiangreco,BartBaesens;UsingSocialNetworkDatatoPredictTechnologyAcceptance.
ICIS2013.
12(2013).
[2]F.
Liebana-Cabanillasa,R.
Noguerasb,L.
J.
Herrerac,A.
Guillenc;Analysingusertrustinelectronicbankingusingdataminingmethods.
ExpertSystemswithApplications.
40(14),5439–5447(15October2013).
[3]S.
Makki,A.
Mustapha,J.
M.
Kassim,E.
H.
Gharayebeh;EmployingNeuralNetworkandNaiveBayesianClassifierinMiningDataforCarEvaluation.
ICGST,4(2011).
[4]HemantaKumarBhuyan,MaitriMohanty,SmrutiRekhaDas;PrivacyPreservingforFeatureSelectioninDataMiningUsingCentralizedNetwork.
IJCSIInternationalJournalofComputerScience,9(3-2),(May2012)[5]Shu-TingLuo,Chwen-TzengSu,Bor-WenCheng;Developingahybridevaluationprocessforproductacceptability:AnEmpiricalStudyinAutomobileIndustry.
5(7),2708-2715(April2011).
12666AcomparativestudyoncarevaluationforecastbasedondataminingBTAIJ,10(20)2014[6]RoisinMcNaney,JohnVines,DanielRoggen;tec.
Exploringtheacceptabilityofgoogleglassasaneverydayassistivedeviceforpeoplewithparkinson's.
CHI'14ProceedingsoftheSIGCHIConferenceonHumanFactorsinComputingSystems,2551-2554.
(2014).
[7]P.
K.
Yadav,K.
L.
Jaiswal,S.
B.
Patel,D.
P.
Shukla;IntelligentHeartDiseasePredictionModelUsingClassificationAlgorithms.
IJCSMC,2(8),102-107(August2013).
[8]AnujSharma,PrabinKumarPanigrahi;AReviewofFinancialAccountingFraudDetectionbasedonDataMiningTechniques.
ComputerScience.
(2013).
[9]ChimingChang,K.
U.
Leuven,Leuven;AComparisonofClassifiersforIntelligentMachineUsagePrediction.
IntelligentEnvironments(IE),2014InternationalConference.
(July2014).
[10]XiuliLi,RuiZhao,YanXiao;ElectronicCommerceDataMiningusingRoughSetandLogisticRegression.
JournalofMultimedia,9(5),688-693(May2014).
[11]ImranKurtOmurlu,MevlutTure,MustafaUnubol,MerveKatranci,EnginGuney;ComparingPerformancesofLogisticRegression,Classification&RegressionTreesandArtificialNeuralNetworksforPredictingAlbuminuriainType2DiabetesMellitus.
IJSBAR,(2014).
[12]YunpengLi,JieLiu,QiuchenBao,WenxiaoXu,RehanSadiq,YongDeng;AnewmethodofmappingrelationsfromdatabasedonArtificialNeuralNetwork.
(Nov2013).
[13]J.
GauravSawale,Dr.
R.
SunilGupta;UseofArtificialNeuralNetworkinDataMiningForWeatherForecasting.
InternationalJournalOfComputerScienceAndApplications.
6(2),(Apr,2013).
[14]ShamsherBahadurPatel,PramodKumarYadav,Dr.
D.
P.
Shukla.
PredicttheDiagnosisofHeartDiseasePatientsUsingClassificationMiningTechniques.
IOSRJournalofAgricultureandVeterinaryScience(IOSR-JAVS).
61-64(Aug.
2013).
[15]ChunFuLina,Yu-ChuYehb,YuHsinHungc,I.
RayChanga;Dataminingforprovidingapersonalizedlearningpathincreativity:Anapplicationofdecisiontrees.
Computers&Education,68,199–210(October2013).
[16]Shen,Runjie,Yang,Yuanyuan,Shao,Fengfeng;IntelligentBreastCancerPredictionModelUsingDataMiningTechniques.
26-27(Aug.
2014).
[17]M.
A.
H.
Farquad,VadlamaniRavi,S.
BapiRaju;Churnpredictionusingcomprehensiblesupportvectormachine:AnanalyticalCRMapplication.
AppliedSoftComputing.
19,31-40(June2014).
[18]D.
Akay,M.
Kurt;Aneuro-fuzzybasedapproachtoaffectivedesign.
Int.
J.
Adv.
Manufact.
Technol.
,40,425-437.
[19]G.
M.
Alam;CangovernanceandregulatorycontrolensureprivatehighereducationasbusinessorpublicgoodsinBangladeshAfr.
J.
Bus.
Manag.
,3(12),890-906(2009).
[20]M.
Alnoukari,W.
Alhussan;Usingdataminingtechniquesforpredictingfuturecarmarketdemand;DCXcasestudy.
InternationalConferenceonInformationandCommunicationTechnologies:FromTheorytoApplications.
IEEEConference.
(2008).
[21]A.
Arauzo-Azofra,J.
M.
Benitez,J.
L.
Castro;AfeaturesetmeasurebasedonRelief.
Proceedingsofthe5thInternationalConferenceonRecentAdvancesinSoftComputing,104-109(2004).