Business failurepredictionusinghybrid2 case-based reasoning(H2CBR)

We haveinvestigatedbusinessfailureprediction(BFP)byacombinationofdecision-aid,statistical,and artificial intelligencetechniques.ThegoalistoconstructahybridforecastingmethodforBFPbycombin- ing variousoutrankingpreferencefunctionswithcase-basedreasoning(CBR),whoseheartisthe k-nearest neighbor (k-NN) algorithm,andtoempiricallytestthepredictiveperformanceofitsmodules.Thehybrid2 CBR (H2CBR) forecastingmethodwasconstructedbyintegratingsixhybridCBRmodules.Thesehybrid CBR moduleswerebuiltupbycombiningandmodifyingsixoutrankingpreferencefunctionswiththe algorithm of k-NN insideCBR.Atrial-and-erroriterativeprocesswasemployedtoidentifytheoptimal hybrid CBRmoduleoftheH2CBR forecastingsystem.Thepredictionoftheoptimalmoduleisthefinal output oftheH2CBR forecastingmethod.Wehavecomparedthepredictiveperformanceofthesixhybrid CBR modulesinBFPofChineselistedcompanies.Inthisempiricalstudy,theclassicalCBRalgorithm based ontheEuclideanmetric,andthetwoclassicalstatisticalmethodsoflogisticregression(Logit)and multivariate discriminantanalysis(MDA)wereusedasbaselinemodelsforcomparison.Featuresubsets were selectedwiththestepwisemethodofMDA.ThepredictiveperformanceoftheH2CBR systemis promising; themostpreferredhybridCBRforshort-termBFPofChineselistedcompaniesisbasedonthe ranking-order preferencefunction.