首页 > 资料专栏 > 论文 > 经营论文 > 风险管理论文 > MBA毕业论文_于ADASYN_RF的P2P网贷借款人信用风险评估PDF

MBA毕业论文_于ADASYN_RF的P2P网贷借款人信用风险评估PDF

资料大小:1554KB(压缩后)
文档格式:PDF
资料语言:中文版/英文版/日文版
解压密码:m448
更新时间:2021/10/24(发布于云南)

类型:金牌资料
积分:--
推荐:升级会员

   点此下载 ==>> 点击下载文档


文本描述
P2P网络借贷是以互联网为运营媒介的新兴借贷模式,为社会中的资金需求群体提 供了一种新的筹资路径,弥补了传统金融机构的不足,缓解了小微企业和个人融资难的 问题。但P2P网络借贷行业一直没有形成完善的个人征信体系,相关监管法律法规不健 全,在满足广大投资者和筹资者需求同时,P2P网络借贷的信用风险问题慢慢显露。目 前,我国P2P网络借贷行业累计平台数达到6621家,平台数量不断增加的背后,各种 风险问题如大量借款人发生违约,平台出现跑路、倒闭等现象频繁发生,P2P网络借贷 的风险控制及管理问题亟待解决。借款人信用风险是P2P网络借贷平台需重点关注的风 险。因此,本文以借款人信用风险为切入点,融入相关“软信息”构建评估指标体系, 并基于网络借贷非平衡交易数据,提出改进后的信用风险评估模型——ADASYN-RF(自 适应综合过采样-随机森林)模型,为网络借贷中借款人信用风险评估问题贡献微薄之力。 第一步,在阅读了大量相关文献的基础上,对国内外针对网络借贷的文献进行梳理, 了解P2P网络借贷信用风险方面的研究现状。第二步,介绍P2P网络借贷相关概念、运 作模式以及发展现状,并就P2P网络借贷信用风险的形成机制与影响因素进行分析。第 三步,构建包含硬信息和软信息的评估指标体系,并介绍随机森林算法、ADASYN-RF 模型的构建、模型性能评价标准。第四步,实证分析。通过对“人人贷”平台2010年 10月至2015年10月的交易数据进行清洗、处理以及描述性统计分析,构建信用风险评 估模型,验证模型的性能。使用当前比较流行的机器学习方法——BP神经网络进行对 比,进一步验证模型的优越性。并根据指标变量在信用风险评估模型中的重要性评分对 其进行分析。实证结果表明:ADASYN-RF模型除了违约样本(少数类)的准确率和未违 约样本(多数类)的召回率有些许降低外,其它评估指标均优于传统随机森林模型;其中, 违约样本的预测召回率上升了23.3个百分点,F1-score上升了8.8%;模型AUC的值提 高10.7%。因此,总体上改进后的模型(ADASYN-RF)具有较优的分类性能。与BP神经 网络对比发现,无论从模型准确率、各类F1-score、AUC的值还是模型ROC曲线来看, ADASYN-RF模型均较优。因此,ADASYN-RF模型的分类性能较好,更适合用于P2P 网络借贷信用风险评估。在分析指标变量的重要程度时发现,借款信息中各变量的重要 程度排名均在前十,属于重要变量。且“软信息”对评估模型的作用不容小觑,其包含 的变量重要程度几乎处于前十。最后,根据实证结果,从监管、征信体系、信息审批制 度以及提高投资人风险意识四个方面提出防范对策与建议,希望有助于P2P网贷平台的 健康持续发展。 关键词:P2P网络借贷;信用风险评估;随机森林;非平衡数据;ADASYN算法 iii Abstract P2PnetworklendingisanewlendingmodewiththeInternetastheoperationmedium, whichprovidesanewfinancingpathforthefunddemandgroupsinthesociety,makesupfor theshortcomingsoftraditionalfinancialinstitutions,andalleviatesthefinancingdifficulties ofsmallandmicroenterprisesandindividuals.However,theP2Ponlinelendingindustryhas notformedaperfectpersonalcreditreportingsystem,andtherelevantregulatorylawsand regulationsarenotperfect.Whilemeetingtheneedsofinvestorsandfundraisers,thecredit riskproblemofP2Ponlinelendinghasgraduallyemerged.Atpresent,thecumulativenumber ofP2PnetworklendingplatformsinChinahasreached6621.Behindtheincreasingnumber ofplatforms,variousriskproblemssuchasalargenumberofborrowers'default,platform running,bankruptcyandotherphenomenaoccurfrequently,whichmakestheriskcontroland managementproblemsofP2Pnetworklendingurgenttobesolved.Theborrower'screditrisk isariskthatP2Ponlinelendingplatformsshouldfocuson.Therefore,thisarticletakesthe borrower'screditriskasanentrypoint,integratestherelevant"softinformation"tobuildan evaluationindexsystem,andbasedontheonlineloanunbalancedtransactiondata,proposes animprovedcreditriskevaluationmodel——ADASYN-RF(AdaptiveSynthesis Oversampling-RandomForest)model,whichcontributeslittletothecreditriskassessmentof borrowersinonlinelending. Inthefirststep,onthebasisofreadingalotofrelevantliterature,theresearchon InternetlendingathomeandabroadissortedouttounderstandtheresearchstatusofP2P networklendingcreditrisk.Thesecondstepistointroducetherelatedconcepts,operating modesanddevelopmentstatusofP2Ponlinelending,andanalyzetheformationmechanism andinfluencingfactorsofP2Ponlinelendingcreditrisk.Thethirdstepistobuildthe evaluationindexsystemincludinghardinformationandsoftinformation,andintroducethe randomforestalgorithm,theconstructionofADASYN-RFmodel,andthemodel performanceevaluationcriteria.Thefourthstepisempiricalanalysis.Bycleaning,processing anddescriptivestatisticalanalysisofthetransactiondataofthe"Renrendai"platformfrom October2010toOctober2015,acreditriskassessmentmodelwasconstructedtoverifythe performanceofthemodel.BPneuralnetwork,apopularmachinelearningmethod,isusedfor comparisontofurtherverifythesuperiorityofthemodel.Andanalyzetheindicatorvariables accordingtotheirimportancescoresinthecreditriskassessmentmodel.Theempiricalresults showthatexceptfortheaccuracyofdefaultsamples(minorityclass)andtherecallrateof iv non-defaultsamples(majorityclass),theotherevaluationindexesofADASYN-RFmodelare superiortothetraditionalrandomforestmodel.Amongthem,thepredictedrecallrateof defaultsamples(minorityclass)hasincreasedby23.3percentagepoints,F1-scorehas increasedby8.8%,andtheAUCvalueofthemodelhasincreasedby10.7%.Therefore,the improvedmodel(ADASYN-RF)hasbetterclassificationperformance.Therefore,the improvedmodel(ADASYN-RF)hasbetterclassificationperformanceparedwithBP neuralnetwork,ADASYN-RFmodelisbetterintermsofmodelaccuracy,F1score,AUC valueandROCcurve.Therefore,theADASYN-RFmodelhasbetterclassification performanceandismoresuitableforP2Pnetworkloancreditriskassessment.Intheanalysis oftheimportanceofindexvariables,itisfoundthatthevariablesintheindexsystemofloan informationareallthetoptenvariablesofimportance,whichareimportantvariables.Andthe roleof"softinformation"intheevaluationmodelshouldnotbeunderestimated,andthe importanceofitsvariablesisalmostinthetopten.Finally,accordingtotheempiricalresults, thepaperputsforwardcountermeasuresandsuggestionsfromfouraspects:regulatorysystem, creditreportingsystem,informationapprovalsystemandimprovinginvestors'riskawareness, hopingtocontributetothehealthyandsustainabledevelopmentofP2Pnetworklending platform. Keywords:P2PNetworkLending;CreditRiskAssessment;RandomForest; UnbalancedData;ADASYNAlgorithm 湖南科技大学硕士学位论文 目录 摘要............................................................................................................................................i Abstract......................................................................................................................................iii 第1章绪论...............................................................................................................................1 1.1研究背景及意义...........................................................................................................1 1.1.1研究背景.............................................................................................................1 1.1.2研究意义.............................................................................................................1 1.2文献综述.......................................................................................................................2 1.2.1P2P网贷运营模式及监管研究..........................................................................2 1.2.2P2P网贷借款人信用风险评估指标研究..........................................................3 1.2.3P2P网贷借款人信用风险评估方法研究..........................................................5 1.2.4P2P网贷非平衡数据研究..................................................................................5 1.2.5国内外研究现状评述.........................................................................................6 1.3研究内容和研究框架...................................................................................................6 1.3.1研究内容.............................................................................................................6 1.3.2研究框架.............................................................................................................7 1.4主要创新之处...............................................................................................................8 第2章P2P网络借贷理论及信用风险分析......