HybridizingprinciplesofTOPSISwithcase-basedreasoning for businessfailureprediction

Case-basedreasoning(CBR)solvesmanyreal-worldproblemsundertheassumptionthatsimilar observationshavesimilar outputs. Asanimplementationofthisassumptionandinspiredbythe techniquefororderperformancebythesimilaritytoidealsolution(TOPSIS),thispaperproposesanew type ofmultiplecriteriaCBRmethodforbinarybusinessfailureprediction(BFP)withsimilaritiesto positiveandnegativeidealcases(SPNIC).Assumingthatthebinarypredictionofbusinessfailure generatestworesults,i.e.,failureandnon-failure,wesettheprincipleofthisCBRforecastingmethod which istermedasSPNIC-basedCBRasfollows:newobservationsshouldhavethesameoutputas the positiveornegativeidealcasetowhichtheyaremoresimilar.FromtheperspectiveofCBR,the SPNIC-basedCBRforecastingmethodconsistsofR4 processes:retrievingpositiveandnegativeideal cases,reusingsolutionsofidealcasestoforecast,retain cases, andreconstructthecasebase.Asa demonstration,weappliedthismethodtoforecastbusinessfailureinChinawiththreedata representationsofa formerly collected dataset from normaleconomicenvironment and arepresentation of a recently collecteddataset from financial crisis environment. TheresultsindicatethatthisnewCBR forecastingmethodcanproducesignificantlybettershort-termdiscriminatecapabilitythan comparativemethods,exceptforsupportvectormachine,innormaleconomicenvironment;Onthe contrary,itcannotproduceacceptableperformanceinfinancialcrisisenvironment.Furthertopics aboutthismethodarediscussed.