Research On Credit Limits For College Students Based On Xgboost And Complex Networks

Since 2016,many P2P financial platforms have begun to develop college credit consumer products.With the gradual expansion of the scale,college students' credit consumption platform is facing the problem of increasing overdue rate and improper allocation of funds.Therefore,this article focuses on the allocation of credit limits for college students on the platform,and conducts research on the renewal of current credit limits of college students and the first allocation of credit limits for newly registered college students.In the updating strategy of college student users' existing quotas,this article selects Internet history information of users on the P2P as the data mining object.Firstly,it extracts and constructs the basic personal information of college students through the feature engineering method in machine learning.The feature dimensions include the order and order behavior information and the behavior information of the APP scene,etc.Then using the indicators such as concentration,IV information value and Pearson correlation coefficient filter the feature set.Based on the characteristics of feature engineering refinement,the Xgboost algorithm framework is used to establish a prediction model of user overdue probability and a prediction model of future consumption amount regression,which are both accurate and stable to the expected level.Combined with the prediction results of the model,the credit status and consumption demand of college students users are evaluated.It not only divides different user levels for the current college student users of the platform,but also makes an effective credit renewal strategy for the existing credit line.After the update,saving the amount of funds for the platform amounted to 14.18%.Due to the lack of historical transaction records and Internet behavior,the method of credit risk and consumer demand assessment of newly registered college students are quite different from previous studies.The paper introduces the concept of Complex Networks.Taking the existing college student users as a node,and constructing a network by connecting edges between the nodes if the overdue probability or the future consumption amount level is same;Then a link prediction algorithm based on node attribute similarity is proposed to predict the dynamic evolution of newly registered university student users entering the network as a new node.The matching function established by transfer that connection information of new and old users divides the credit rating and consumption demand of the newly registered college students by using the information transmitted by the old users,and formulates the first allocation strategy of the credit line of the newly registered college students.At last,the real user data of college students is selected to demo the operation of the model in practical application,in which the accuracy of the link prediction model is 0.68,reaching the expected standard.This article conducts a comprehensive study on the credit limits of university students.At the same time,it conducts an experimental verification for the credit lines updating and credit lines allocation algorithm,which achieves a better optimization and prediction effect.It is significant to the entire Internet financial platform.