Application Of CNN Model Based On XGBoost Feature Screening In Customer Credit Evaluation

Nowadays,the development of the economy and the renewal of the ways of payment have changed people’s concept of consumption,and the way of spending before repaying is gradually accepted and rapidly becoming popular.From company financing to people buying houses and cars,from paying for furniture and electric appliances by credit card installments to using AntPay and Jingdong White Stripe for household goods,lending is a part of every aspect of life.It is important for lending institutions to make accurate judgments about whether their customers can repay their loans properly.On the one hand,if a lending institution approves a loan to a customer under poor credit circumstances due to an error in judgment,the failure of the customer to repay the loan as promised can result in financial loss to the lending institution.On the other hand,if a lending institution refuses to grant a loan to the customer under good credit circumstances,the reputation of the lending institution can be damaged.Therefore,it is particularly important to evaluate the credibility of lenders,to distinguish as accurately as possible between normal users and defaulters before a loan occurs,and to reduce the financial and credit losses caused by misclassification of loan customers.This paper introduces the background and significance of credit assessment re-search in the first part,analyzes the importance of credit assessment,and provides a preliminary understanding of the development of credit assessment with the rel-evant literature on credit assessment at home and abroad.In the second part,we focus on the feature screening approach based on the XGBoost algorithm and the Convolutional Neural Network classification model.In the third part,by analyzing the data of foreign lending company Lending Club of 2019,we have verified that the Convolutional Neural Network model based on XGBoost feature screening is high-quality in the customer credit assessment problem.In terms of empirical ev-idence,the preliminary dataset has been obtained by pre-processing the variables on the original dataset.Then using the XGBoost algorithm to feature screen the remaining variables,and using two feature screening methods,Ⅳ value and princi-pal component analysis,as a comparison,through the convolutional neural network to verify the superiority of feature filtering,which means that using XGBoost for feature screening improved the accuracy of model.To verify the effectiveness of Convolutional Neural Networks in customer credit evaluation,the features screened by XGBoost was used as input variables to compare Convolutional Neural Networks and four other traditional classification models.Then we have analysed the effec-tiveness of the models by evaluating both the ability to differentiate between positive and negative samples and the overall performance of the models.We have found that the Convolutional Neural Network model achieves a high level of differentiation between normal and defaulted users as well as the overall performance of the model,and the Sigmoid activation function set in the fully connected layer of the Convo-lutional Neural Network makes the output result the probability of default,which makes the model built by combining XGBoost algorithm and Convolutional Neural Network not only has high classification accuracy,but also can determine whether or not to approve a loan by using the probability of default based on the institution’s actual situation in practice.Therefore,the Convolutional Neural Network model based on the XGBoost algorithm is more suitable for the evaluation of customer credit problems.In the fourth part of this paper,we have summarized the research content of the whole paper and gives rationalized suggestions and prospects for the development of customer credit assessment in China.In this paper,we have used XGBoost algorithm and Convolutional Neural Net-work model to study customer credit problems,and verified that this method can improve the classification accuracy of the credit assessment model and help lending institutions to make decisions on whether to approve loans for new customers based on the classification results and their actual situation.The content of this paper can reduce customer credit risk to a certain extent,reduce the financial losses and improve the operational efficiency of lending institutions,and improve the customer credit assessment system theoretically.