Research On Loan Risk Assessment Method Integrating Consumption Data

As an important part of the development of Online finance,loan risk assessment is one of the hot topics in the research of Online finance.And with the rapid development of the Internet technology,users have generated a lot of data on the Internet.Through data mining and analysis,and converts it to reflects the characteristics of the user's loan risk,which can describe loan users from different perspectives.By increasing the dimensions of loan risk assessment,the ability of loan risk assessment of Online finance is improved and the bad debt rate is reduced at the same time.In this thesis,the user's consumption data on the consumption platform is introduced to evaluate the user's loan risk,and the research is carried out from the dimension of loan risk assessment and loan risk assessment model.The main research of this thesis are as follows:1.After preprocessing the consumption data,user consumption profile characteristics are constructed from the four dimensions including user consumption level,consumption frequency,consumption stability and potential problems,and the clustering variables generated by clustering are also added to the consumption profile characteristics.On the basis of using user's basic information features,this thesis adds consumer portrait features to evaluate user's loan risk,and compares the results of three feature sets,including user's basic information features,consumer portrait features and features combining user's basic information and consumer portrait,in four groups of machine learning models.And conduct 20 times of 10-fold cross-validation,use statistical learning method to analyze sets of result sample data,and the result shows that adding the consumption portrait feature improves the classification result of the loan risk assessment model and can reduce the bad debt rate.2.Use deep learning to evaluate loan risk based on sequence features in user consumption data in this thesis.Referring to the text emotion classification task,each consumption is compared with the words in emotion classification.By transforming the consumption data into consumption vector and inputting it into the deep learning model,this thesis compares the influence of different model structures on the classification,and considers that each consumption has different weights for the evaluation results,attention mechanism is introduced.The experimental results show that using consumption sequence can distinguish high-risk and low-risk users in Online financial loans.3.This thesis constructs a loan risk assessment model integrating consumption data.First,the basic information characteristics of users,the consumption portrait characteristics and the consumption sequence characteristics are processed separately,then input to the network layer for feature fusion,and finally output the loan evaluation results.The experiment compares the single model,the integrated model and the fusion model built in this thesis.The results show that the evaluation results of the fusion model are better than the single model and the integrated model in the two evaluation,which can better distinguish the high-risk and low-risk users and reduce the bad debt rate.