【作者】 孙伟；

【导师】 王星惠；

【作者基本信息】 安徽大学 ， 应用统计硕士（专业学位）， 2019， 硕士

【摘要】 近些年来,随着中国互联网的不断发展,互联网金融和电子商务也得到前所未有的发展,互联网金融越来越被人们重视。P2P网络贷款是现在互联网金融的主要模式,作为第三方中介自身不提供资金和贷款,只提供平台并且对参与者的身份和信用情况进行审核,促进借款贷双方在平台完成交易。P2P网贷为小微企业和个人提供了小额信贷的可能,不仅将社会上的闲置资金充分的利用起来,而且还给有资金需求的人带来了方便。2007年中国首家P2P网贷平台成立,从此中国的传统金融模式有所改变。随后几年里中国出现大量的P2P网贷平台,平台的数量越来越多,但是其中也存在着大量的问题,比如平台风险预测和掌控体系不完善、借款利率低以及借款成功的用户占比低等,本文着重研究的网贷平台是人人贷,数据是来自人人贷官网公布的散标数据。本文研究的是网络贷款平台借款的成功性,P2P网贷平台借款失败(流标)是指借款人在平台发布借款的申请之后,规定的时间内没有足够的贷款人愿意投标进而不能筹集到足额的资金。是否会流标不仅关系到借方能否筹集到自已所需的资金、贷方能否将自已的闲置资金贷出并且从中取得利益,而且还关系到平台的运营和收益情况。经济活动中,参与交易的双方在获得信息的渠道和质量上都是不同的,交易过程中的关键信息往往掌握在借款人手中,即信息是不对称的,所以本文研究借款人的信息与借款成功性之间的关系。通过查阅与分析国内外的相关资料和参考文献,试图从借款人提交的个人信息出发,来探究借款人在平台上借款成功与否的过程,同时这也是投资者对于借款人进行筛选的过程,对是否流标进行预测,通过研究结果来帮助借款人如何在平台上提高借款的成功率,帮助投资者识别存在风险的借款人,降低投资者的风险。首先挑选适合的特征,并将特征总结成4类。其次删除数据中的缺失值和异常值,并对数据实施转换,使之适合模型的输入要求。然后对数据进行描述性统计,探究出每个特征对于最终是否借款成功的影响情况,并探究其中的原因。然后将数据的2/3当作训练集,剩余数据作为测试集,借助逻辑回归、支持向量机和人工神经网络对数据进行训练与预测。对于模型的分类效果,使用准确率、召回率、ROC曲线和AUC值来进行评价,结果是逻辑回归不仅在训练集上的拟合效果很好,而且预测效果也很理想,说明通过挑选出的特征和模型能够很好的预测P2P网贷平台的借款成功性情况。同时对借款成功的影响因素进行了探讨,借款金额、借款利率、借款期限、是否有房贷、是否进行身份认证等都是显著的影响因素,借款人鉴于此可以降低借款失败的可能性、成功筹集资金,同时对投资者来说,可以帮助他进行投资决策。更多还原

【Abstract】 In recent years,with the continuous development of China’s Internet,Internet finance and e-commerce have also been unprecedentedly developed,and Internet finance has been paid more and more attention.P2P network loans are the main mode of Internet finance now.As a third-party intermediary,it does not provide funds and loans.It only provides a platform and reviews the identity and credit status of participants,and promotes the borrowing and lending parties to complete transactions on the platform.P2P online lending provides the possibility of microfinance for small and micro enterprises and individuals.It not only makes full use of the idle funds in the society,but also brings convenience to those who have financial needs.In 2007,China’s first P2P online lending platfonn was established.Since then,China’s traditional financial model has changed.In the following years,there were a large number of P2P online lending platforms in China,and the number of platforms increased.However,there were also a lot of problems,such as platform risk prediction and imperfect control system,low borrowing rate and the proportion of users who borrowed successfully.Inferior,the online lending platform that this article focuses on is everyone’s loan,and the data is from the scattered data published by Renren.This paper studies the success of the loan on the online loan platform.The failure of the P2P online loan platf-orm means that after the borrower has applied for the loan on the platform,there are not enough lenders who are willing to bid and cannot raise the full amount within the prescribed time.Funds.Whether or not the flow label will not only affect whether the borrower can raise the funds it needs,whether the lender can lend its own idle funds and obtain benefits from it,but also the operation and income of the platform.In economic activities,the parties involved in the transaction are different in the channels and quality of information.The key information in the transaction process is of-ten in the hands of the borrower,that is,the information is asymmetric,so this paper studies the borrower’s information and borrowing.The relationship between success.By reviewing and analyzing relevant data and references at home and abroad,we try to explore the process of borrowing the borrower’s success on the platform from the personal information submitted by the borrower.At the same time,this is also the process of investors screening the borrower.Forecast whether the flow label is used,and help the borrower to improve the success rate of borrowing on the platform through the research results,help investors identify the risky borrowers,and reduce the risk of investors.First select the appropriate features and summarize the features into 4 categories.Secondly,the missing values and outliers in the data are deleted,and the data is transformed to fit the input requirements of the model.Then descriptive statistics on the data to explore the impact of each feature on the ultimate success of the loan,and explore the reasons.Then 2/3 of the data is taken as the training set,and the remaining data is used as the test set.The data is trained and predicted by means of logistic regression,support vector machine and artificial neural network.For the classification effect of the model,the accuracy,recall rate,ROC curve and AUC value are used for evaluation.The result is that the logistic regression not only has a good fitting effect on the training set,but also the prediction effect is very satisfactory,indicating that the selected results are selected.Features and models can well predict the success of borrowing on P2P online lending platforms.At the same time,the factors affecting the success of the loan are discussed.The borrowing amount,the borrowing rate,the borrowing period,whether there is a mortgage,and whether the identity is authenticated are all significant factors.In view of this,the borrower can reduce the possibility of borrowing failure and successfully raise funds.Funding,while for investors,can help him make investment decisions.更多还原