Research On Personal Credit Risk Assessment Using Logistic Regression And BP Neural Network Models Based On The LendingClub Dataset

In recent years,with the continuous improvement of residents’ income level in China,the consumption volume of residents has also significantly increased,and at the same time,personal credit business is also constantly increasing.With the rapid expansion of personal credit business,the assessment of personal credit risk becomes particularly important.First,the article uses the data set provided by Lending Club to make a descriptive analysis of the original data,and makes a preliminary assessment of personal credit risk from the borrower’s working hours,loan term,loan amount,loan interest rate,borrower’s occupation,credit rating,borrower’s housing status and other factors.Secondly,preprocess the data and construct necessary new features;The classification features are combined with statistical indicators for manual sub box processing,the continuous feature variables are subject to Chi-squared test based on IV values,and the optimal sub box processing is carried out according to the test results;Convert data into WOE value representation;Then,personal credit risk assessment models are constructed based on Logistic regression algorithm and BP neural network model respectively.The performance of the Logistic regression model is tested by using Confusion matrix,Receiver operating characteristic and other classifier performance evaluation indicators,and the results show that the AUC value on the test set data is 0.921,which indicates that the Logistic regression model constructed in this paper has good stability and classification effect on personal credit risk assessment;Test and compare the BP neural network model of traditional indicators and the BP neural network model of improved indicators from the aspects of prediction accuracy,Learning rate and iteration times.Finally,by comparing the Logistic regression model with the BP neural network model from the accuracy rate,recall rate and F1 score,the following conclusions are drawn: as a pre loan risk control model,the Logistic regression model is widely used in commercial banks;In the Logistic regression model,multiple factors such as the user’s historical borrowing records,income,and working hours can be taken into account by assigning weights to the evaluation indicators,so as to better evaluate the user’s credit risk rating.The BP neural network model can also optimize the model structure and parameters through the Backpropagation to improve the prediction ability of the model.Regardless of which model is used,qualitative and quantitative analysis of users’ credit risk requires considering multiple factors,including historical loan records,income situation,working hours,etc.,in order to expand the business volume of financial institutions to a greater extent while minimizing their credit risk,thereby promoting the development of the financial industry to a greater extent.This study has certain reference value for lending institutions to effectively avoid risks,better manage and operate lending businesses.Finally,based on the evaluation of personal credit risk models,several prevention suggestions are proposed,including enhancing personal legal awareness and risk prevention awareness;Increase pre loan audit indicators and evaluation dimensions;Introduce various mathematical algorithm models and financial technology to strengthen personal credit risk management;Establish a comprehensive system of personal credit risk assessment indicators.