基于人工神经网络算法的信用风险预测(IJMECS-V10-N5-2)

I.J.ModernEducationandComputerScience,2018,5,9-16PublishedOnlineMay2018inMECS()DOI:10.5815/ijmecs.2018.05.02Copyright©2018MECSI.J.ModernEducationandComputerScience,2018,5,9-16CreditRiskPredictionUsingArtificialNeuralNetworkAlgorithmDeepakKumarGuptaUCDMichaelSmurfitGraduateBusinessSchool,Dublin,IrelandEmail:[e-mail]murfitGraduateBusinessSchool,Dublin,IrelandEmail:[e-mail]:10March2018;Accepted:10April2018;Published:08May2018Abstract—Artificialneuralnetworkisaninformationprocessingsystemwhichisinfluencedbythehumanbrainandworksonthesameprinciplesofthebiologicalnervoussystem.Theypossesstheabilitytoextractmeaningfromcomplexandintricatedata,bydetectingtrendsandextractingpatternsfromit.Thispaperillustratestheabilityofneuralnetworkmodelandlinearregressionmodelconstructedtopredictthecreditworthinessofanapplicationaccuratelyandpreciselywithminimalfalsepredictionsanderrors.Theresultsareshowntobesimilarforboththemodels,thus,modelsareefficienttousedependingonthetypeofapplicationandattributes.IndexTerms—CreditRisk,ArtificialNeuralNetwork,LinearRegression,CreditRiskAnalysis,CreditRating,CreditRatingI.INTRODUCTIONCreditriskorcreditdefaultindicatestheprobabilityofnon-repaymentofbankfinancialservicesthathavebeengiventothecustomers[1].Creditriskhasalwaysbeenanextensivelystudiedareainbanklendingdecisions.Creditriskplaysacrucialroleforbanksandfinancialinstitutions,especiallyforcommercialbanksanditisalwaysdifficulttointerpretandmanage[2].Duetotheadvancementsintechnology,bankshavemanagedtoreducethecosts,todeveloprobustandsophisticatedsystemsandmodelstopredictandmanagecreditrisk.Theobjectiveofcreditriskmodelsistoevaluatetheriskportfoliooftheborrowerandthenassignaprobabilityofdefault[3].Therefore,therehasbeenadiscussiononclassificationanddiscriminationproblemsforsolvingcreditriskmodels[4].Banksevaluateloanapplicationsbasedonasubjectiveassessmentmadebytheborrower[5].Thisassessmentcanleadtoinefficientandinconsistentapplications.Bankswillbesuccessfuliftheyareabletoreducethecreditriskandhaveasignificanteffectoneconomicgrowthofthecountry.Todiscriminatebetweengoodcustomersandbadcustomers,banksdevelopedaneedforamodel-basedapproachthatcanpredictcreditdefaultaccurately[6].Themodel-basedapproachprovidesbettercreditdefaultmanagementandefficientlyallocatecapital[7].Topredictthecreditdefault,severalmethodshavebeencreatedandproposed.Theuseofmethoddependsonthecomplexityofbanksandfinancialinstitutions,sizeandtypeoftheloan[8].Thecommonlyusedmethodhasbeendiscriminationanalysis[9].Thismethodusesascorefunctionthathelpsindecisionmakingwhereassomeresearchershavestateddoubtsonthevalidityofdiscriminatesanalysisbecauseofitsrestrictiveassumptions;normalityandindependenceamongvariables[10].Artificialneuralnetworkmodelshavecreatedtoovercometheshortcomingsofotherinefficientcreditdefaultmodels[11].Theobjectiveofthispaperistostudytheabilityofneuralnetworkalgorithmstotackletheproblemofpredictingcreditdefault,thatmeasuresthecreditworthinessoftheloanapplicationoveratime[12].Feedforwardneuralnetworkalgorithmisappliedtoasmalldatasetofresidentialmortgagesapplicationsofabanktopredictthecreditdefault[13].Theoutputofthemodelwillgenerateabinaryvaluethatcanbeusedasaclassifierthatwillhelpbankstoidentifywhethertheborrowerwilldefaultornotdefault.Thispaperwillfollowanempiricalapproachwhichwilldiscusstwobasedonneuralnetworkmodelsandexperimentalresultswillbereportedbytrainingandvalidatingthemodelsonresidentialmortgageloanapplications[14].Asthefinalstepinthedirection,linearregressionmethodisalsoperformedonthedataset.Resultswillprovideacomparisonbetweentheefficiencyandaccuracyoftheneuralnetworkandlinearregressionmethods[15].Asthepaperfollowsanempiricalapproach,thispaperwillshowstructuredexperimentalapproachtothedesignofmodels.II.LITERATUREREVIEWTheprimaryobjectiveofcreditevaluationprocessistocomparecharacteristicsofanapplicantwithotherpreviouscandidateswhohaverepaidtheloanamount.10CreditRiskPredictionUsingArtificialNeuralNetworkAlgorithmCopyright©2018MECSI.J.ModernEducationandComputerScience,2018,5,9-16Bankwillcheckcandidate'sprofilewithearliercandidates,ifaprofileisverymuchsimilar,thentheywillcheckifanapplicanthasrepaidtheloanontime[16].Ifaclaimantdidnotdefaultthentheloancanbegranted,ifnotthenloanapplicationwillberejected.Twotechniquesforcreditevaluation:CreditScoringandOfficialsSubjectiveAssessment.Traditionaljudgementassessmentmethodisentirelydependentonevaluator'sexperienceandknowledge[17].Subjectiveassessmentissubjectiveandinconsistent,butontheotherhanditcanbesuccessful,creditor'sexperiencecanbequalitativethathelpsintakingsuccessfulcreditdecisions[18].Whileincreditscoringmethod,creditorsusetheirknowledgeandhistoricalinformationoftheloanapplicationstoformanevaluationmodeltodeterminecreditworthiness[19].Creditscoringmethodsareconsistent,andself-operatedthatincludesquantitativemeasurementsofapplicant'screditscoresubjectedtopredictorvariablessuchasemploymentdurationorcredithistory.Also,cred