Planning Of Network Loan Default Risk Assessment Scheme Based On Convolutional Neural Network And Survival Analysis

Online loans are an extension of traditional loans online,but there is also a risk of default in online loans.Offline loan risk assessment methods are not applicable to online loans.This paper analyzes and models the data of Lending Club from 2007 to 2018,and builds a corresponding network loan default risk assessment model through convolutional neural networks and survival analysis models.The convolutional neural network is used To predict the default probability of online loans,a survival analysis model is used to predict the possible default time.After data processing,97 variables were obtained from the 151 feature variables in the data set to establish a default risk assessment model,and the prediction results were compared with the prediction results of machine learning models commonly used in the industry.The comparison found that the convolutional neural network model has a prediction accuracy of 83.75% for defaulted loans,while the Logistic model has a prediction accuracy of 80.23% for online loan defaults,and the random forest's network loan default prediction accuracy rate is 76.68%.The accuracy rate of online loan default prediction is 70.46%.The results show that the convolutional neural network model of network loan default prediction accuracy is higher than other traditional machine learning methods,and can more accurately assess the default risk of network loan borrowers.Based on previous studies,12 feature variables were selected and Cox regression model was used to predict the default time of network loans.The research results show that the survival curve of online loans shows a downward trend,that is,the survival probability of online loans is inversely related to time,which is also consistent with the perception of reality,that is,as the repayment time increases,the borrower is more likely The loan cannot be repaid.And we can see that the survival probability decreases rapidly in the last period,that is,close to 60 months.After verification,the accuracy of the default probability prediction model built in this paper can reach 83.53%,which has a good prediction effect.Combining a convolutional neural network model for predicting default probability and a survival analysis model for predicting default time,forming a framework for evaluating the default of online loans,and providing risk management recommendations for online loan platforms so that they can better carry out risk Management for sound business.