Modeling Design And Implementation Of Overdue Behaviors On User's First Loan

Loan is one of the most important attributes of financial business.With the explosive growth of loan business,the overdue rate of borrowers continues to rise.For the first-time loan business,it is difficult for financial institutions to determine whether the borrower will repay overdue when approving the loan,which increases the difficulty of loan audit to a certain extent.How to accurately measure the qualification of first-time loan users to reduce the overdue risk has become an important problem to be solved by financial institutions.Based on the traditional machine learning and deep learning algorithms,this paper studies the first loan overdue behavior of users,and provides the corresponding solutions.At present,the data sets of common overdue prediction models are mostly the data generated by the borrower after the loan,and there is little research on the user characteristics left by the borrower on the platform before the loan.Therefore,in order to further explore the loan characteristics,this paper selects the user's pre loan and post loan data as the research object.Data set a(post loan)is the loan data of an online product on Dr company's credit platform,and data set B(pre loan)is the borrower's loan application record on the platform in data set a(data by the end of2018).By analyzing the data characteristics of users before and after loan,a feature set construction method based on feature grouping reconstruction is proposed,and a combined feature prediction model based on xgboost DNN is designed,and then the model is used to predict whether the overdue behavior of first-time loan users occurs,and realized a on-line predictions system with the proposed model,details as follows:(1)Feature set construction method based on feature grouping reconstruction:firstly,the characteristics of the two data sets are statistically analyzed.Through the analysis,taking each natural day as an independent observation day,the data are grouped according to the borrower's marriage attribute,city attribute,occupation attribute and loan amount attribute to construct different types of loan approval pass rates In the approval process,there are five new features: user cancellation rate,30 day overdue rate,7-day overdue rate and incoming reminder rate.Then,the IV value of the features under each group is calculated through woe code to evaluate the prediction performance of the new features,and the optimal feature subset is selected in combination with xgboost feature selection algorithm.(2)Based on xgboost DNN combined feature prediction model: firstly,based on the results of feature engineering construction,random forest and xgboost are selected for single model training.Secondly,grid search algorithm and k-fold cross validation are used to optimize the single model,analyze the feature importance and group input to xgboost DNN combined feature prediction model for modeling.Through the accuracy of the model The accuracy,recall,F1 value and AUC value are used to compare and evaluate the performance of the model.The experimental results show that xgboost DNN combined feature prediction model has higher prediction accuracy and stability than single models such as random forest and xgboost,which proves the effectiveness of this method.(3)Designed and realized a risk control platform of the user's first overdue prediction system with XGBoost-DNN combined prediction model as the core algorithm,witch servers as the loan approval operation platform for the enterprise credit approval officers.