神經(jīng)計算及其在財務失敗預測中的應用.pdf

1、北京交通大學碩士學位論文神經(jīng)計算及其在財務失敗預測中的應用姓名：王靜申請學位級別：碩士專業(yè)：信號與信息處理指導教師：裘正定20040301北京交通大學碩士學位論文AbstractWiththedevelopmentofmarketeconomyresearchonfinancialfailurepredictionbecomestheurgentrequirementofbanksbusinesscirclesandgovernment

2、managementorganizations.AfteryearsofdevelopmenttherecomesaclimaxofresearchonthepredictionofcorporatefinancialfailureinWesterncountries.Recentyearsmanynewmodelsandmethodsinpatternrecognitionandmachinelearningfieldshavebee

3、nintroducedintothisnewfieldandhavechallengedpopulartraditionalstatisticmodels.HoweverrelatedresearchworkjustbeginsandmainlylimitstousingtraditionalmodelsinChinaComparedwithtraditionalstatisticmodelsusedinfinancialfailure

4、predictionresearchneuralnetworkshavethepowerofgrasping，mplandC0nonlinearrelationshipstrongselfadaptationandgeneralizationabilitybecomeanewfocusontheresearchfieldinforeigncountries.eXsUThispaperregardsChinesepubliccompani

5、esreceivedspecialtreatmentasasignaloffinancialfailure.AfterproperchoosingdatasetsandfeatureselectionwesuccessfullyestablishedpredictingmodelswithBackPropagation(BP)neuralnetworksandlearningvectorquantization(LVQ)neuralne

6、tworks.ComparedwithC4.5andLogisticmodelsthesetwoneuralnetworksshowbeterpredictingresultsespeciallytheLVQmodelsInordertoimprovethegeneralizationofneuralnetworkmodelsitoftenneedsmanytimesoftrialstofindthebeststructureofneu

7、ralnetworks.Neuralnetworkensemblecanavoidthisproblemandalsoimprovethegeneralizationthroughtrainingseveralneuralnetworkssolelyandintegratingtheiroutputs.ThispaperusingtheensembleapproachofBaggingtoconstructthefinancialfai

8、lurepredictionmodelandtheexperimentresultsshowneuralnetworkensembleoutperformssingleneuralnetworksThisstudydesignsandrealizesapracticalfinancialanalysissoftwaresystemwhichrealizedpreliminaryfinancialanalysisfunctions.The

9、systemprovidedadataplatformforthispapersresearchwork.ItalsoprovidesabasisfortheembedmentandpracticalutilizationofthemodelsstudiedinthispaperKeywordsneuralcomputationartificialneuralnetworkfinancialfailurepredictionBPLVQn