摘要
Creditriskpredictionmodelsseektopredictqualityfactorssuchaswhetheranindividualwilldefault(badapplicant)onaloanornot(goodapplicant).Thiscanbetreatedasakindofmachinelearning(ML)problem.Recently,theuseofMLalgorithmshasproventobeofgreatpracticalvalueinsolvingavarietyofriskproblemsincludingcreditriskprediction.OneofthemostactiveareasofrecentresearchinMLhasbeentheuseofensemble(combining)classifiers.Researchindicatesthatensembleindividualclassifiersleadtoasignificantimprovementinclassificationperformancebyhavingthemvoteforthemostpopularclass.Thispaperexploresthepredictedbehaviouroffiveclassifiersfordifferenttypesofnoiseintermsofcreditriskpredictionaccuracy,andhowcouldsuchaccuracybeimprovedbyusingpairsofclassifierensembles.Benchmarkingresultsonfivecreditdatasetsandcomparisonwiththeperformanceofeachindividualclassifieronpredictiveaccuracyatvariousattributenoiselevelsarepresented.Theexperimentalevaluationshowsthattheensembleofclassifierstechniquehasthepotentialtoimprovepredictionaccuracy.
出版日期
2009年03月13日(中国期刊网平台首次上网日期,不代表论文的发表时间)