Design And Implementation Of Credit Rating System Using XGBoost Algorithm

Since the subprime mortgage crisis in the United States in 2007 and the release of international regulations in the Basel III on September 12,2010,credit risk management has become a major concern for banks.In order to prevent credit fraud and distinguish customers with high credit risk from customers with low credit risk,banks need to use credit scoring models to predict credit risk,thereby avoiding economic losses.The scientific credit scoring process is a combination of quantitative analysis and qualitative judgment.Based on the prediction results of the credit scoring model,banks decide to approve or reject the loan application.A good credit scoring model can save unnecessary costs for banks,improve work quality,and reduce credit risk.The purpose of this article is to design and implement a bank credit rating system that facilitates banks to view basic customer information,loan information,and credit scores.The focus of this system is credit scoring.Banks need to use the credit scoring model to predict the probability of a customer's default according to the customer's relevant information,and then convert the probability into credit scores.Banks use the system's credit scores to classify customers.Based on the scoring results,banks decide whether to approve a loan application from the customer.The establishment of this system can not only improve the bank's loan approval efficiency,but also reduce the bank's manpower and material costs.The main work of this paper includes the following two aspects:(1)Establish a credit scoring model.In this paper,first of all,through literature analysis,I learned about the credit scoring models that are currently used and have good prediction effects,and the main solutions for imbalanced data processing.A method of combining the SMOTE algorithm and the XGBoost algorithm to build a credit scoring model was proposed.In the process of building the model,because the collected data set has missing value and abnormal value and the class is not balanced,the first step is to deal with the missing value and abnormal value of the data set,for the unbalanced data set SMOTE resampling.The next step is feature selection.Next,establish logistic regression,random forest,and XGBoost models respectively,and use the ROC curve and AUC to compare the classification effect of the evaluation indicators commonly used in classification models.Finally,XGBoost,which has better performance and prediction effect,is selected as a credit scoring model and imported into the credit rating system developed in this paper to achieve the credit scoring of customers.(2)Design and implement a credit rating system.First of all,this article introduces the system development technology.This system is based on the Python programming language,Bootstrap front-end development technology,and Django development framework.The data store is a MySQL database.Secondly,this article introduces the demand analysis and system design of the credit rating system,including the overall architecture design of the system,the design of each functional module and the design of the database.Finally,the implementation part of the system is introduced according to the division of functional modules.The credit rating system developed in this article has four functional modules,including user management function,credit scoring function,risk early warning function,and excellent customer management function.