Font Size: a A A

A Hybrid Credit Scoring Model Based On BP-PSO-AdaBoost Algorithm And Its Application In Peer-to-Peer Lending

Posted on:2019-08-26Degree:MasterType:Thesis
Country:ChinaCandidate:X C ZhaoFull Text:PDF
GTID:2439330563496626Subject:Finance
Abstract/Summary:PDF Full Text Request
With the rapid development of credit economy in China,the proportion of consumer credit and personal loan business in financial institutions also increased substantially.Furthermore,peer-to-peer lending has become an important form of inclusive finance in our country with respect to the rapid development of information technology.According to the statistics of WDZJ.com,a total of 5,970 P2P lending platforms have been operated in China by the end of November,2017.And the total lending amount of the entire P2P online lending industry reached 6.01 trillion yuan in twelve years,and the whole industry lent 2.8244 trillion yuan in 2017.With the help of information science and big data technology,P2P lending broke the limitations of the traditional relationship between borrower and lender,and effectively reduced the financing cost of borrowers,and raised the financing efficiency of social funds and greatly supplemented the existing financial system in our country.However,with respect to the fictitiousness of Internet information,the phenomenon of information asymmetry is still serious in P2P lending.In addition,compared with some developed countries aboard,the construction of China's credit system is still relatively backward.The credit risk of the borrower becomes more difficult to prevent in the P2P lending,which also severely restricts the development of the P2P lending industry.Therefore,how to evaluate the borrower's credit risk scientifically and precisely and thereby establish an effective personal credit risk assessment model is an urgent problem to be solved for the P2P online lending industry in our country.This study introduced the concept of P2P lending,loan process and credit risk features briefly in section 2;then we analyzed the credit risk in P2P lending in terms of the borrower,the investor and the guarantee agencies.Therefore,the construction of the theoretical basis for our personal credit risk assessment model is built.Secondly,in an effort to construct the personal credit evaluation index system,this study reviewed the relevant research literature of credit evaluation worldwide.According to the credit rating system of financial institutions both at home and abroad,and we also fully considered the personal credit risk characteristics of P2P lending.We finally constructed our credit rating index system based on four parts,which includes the basic personal information,loan information,credit information and personal related certification information.Thirdly,we redefined the values of the variables in our model according to the credit rating index system built above,then a method of SMOTE algorithm was adopted to rebalance the unbalanced credit dataset;In this study,BP neural network is selected as the basic classifier for our hybrid model,and we also used the particle swarm optimization(PSO)algorithm to optimize the initial weights and thresholds of the basis classifier.Then the Ada Boost algorithm framework is used to integrate the optimized basic classifier.For the purpose of evaluating the hybrid model performance,we selected a group of evaluation metrics includes G-Mean,F-Measure,ROC / AUC and Accuracy.Finally,we evaluated the performance of our hybrid classification model over a real-world credit dataset.We compared the classification results of our model with that of other seven traditional classifiers,the proposed model is much superior to the traditional credit scoring model over all the metrics.Therefore,our hybrid credit risk assessment model is much more suitable to evaluate the credit risk of borrowers in P2P lending.Finally,some concrete policy suggestions in promoting the healthy development of P2P lending in China were given in section 5.
Keywords/Search Tags:Peer-to-peer Lending, Personal Credit Risk, BP Neural Network, Particle Swarm Optimization, AdaBoost Algorithm
PDF Full Text Request
Related items