Credit Risk Assessment Based On CatBoost Fusion Algorithm Evaluation And Model Research

With the rapid development of economy,people begin to consume in advance through bank credit card,Huabei,Jingdong Baitiao and other new loan products.In this context,China's consumer credit loan is constantly developing.But there are many uncertain factors in the credit business itself,which will bring about credit risk.So it is particularly important to introduce more effective credit evaluation methods to evaluate the personal credit risk of banks.Accurate and efficient credit evaluation model can improve the ability of risk prediction and play an effective role in preventing credit loan risk.Therefore,we study the performance of integrated algorithm based on credit data to provide credit scoring model for commercial banks.In this paper,first of all,through reading the research literature,learning the previous scholars' research on the credit risk model,to understand the history of the gradual development of credit scoring card,the relevant theory of credit risk assessment,and the principle knowledge of each integration algorithm.Secondly,the variables are selected by Lasso,and then a new Lasso catboost fusion model algorithm is proposed based on catboost algorithm in the integration algorithm.The model is established based on a credit data set in the kaggle network,and compared with a variety of integration algorithm models: random forest,Ada Boost,xgboost,lightgbm and logistic regression algorithm.Empirical analysis shows that the lasso catboost algorithm proposed in this paper has the best classification performance under the comprehensive consideration of each evaluation index.Finally,this paper establishes a score card model based on the logistic regression algorithm which has the advantages of strong interpretability and easy operation.According to the rules of ? value,the characteristic variables are selected to establish a logistic regression model,and then the influencing factors are scored through the weight of evidence transformation(woe),so as to establish a score card model and get the total score of each customer.In the actual personal credit business of commercial banks and other enterprises,we can establish a credit scoring card suitable for our own enterprises according to the actual credit data,grade the applicants,and finally make decisions such as whether to borrow or not.Due to the accurate classification effect and strong explanatory ability of logistic algorithm model,it is still worth studying in personal credit risk assessment model.