Research On Scoring Model Of Bank Credit Applicants Based On Machine Learning

With the vigorous development of the Internet financial industry,banks and other financial institutions are also taking advantage of the wave of Internet development to expand their business and customer scope.With the continuous development of banks and other financial institutions,more and more financial fraud problems begin to appear.Therefore,banks and financial institutions began to pay attention to the problem of fraud in credit applications.How to distinguish a credit applicant is a worthy research topic.This topic is in the context of the current big data,aiming at the shortcomings of the existing applicant scoring model of banks and other financial institutions.At present,the applicant scoring model widely used in banks and other financial institutions is based on the logistic regression algorithm.But the accuracy of the logistic regression model is not very high,and the form is relatively simple,which can not fit the actual application data well.In view of the above shortcomings,this paper mainly does the following work:1.Data processing.The data set of this experiment covers 250000 records including users’ personal financial status.It includes 150000 labeled data and 100000 unlabeled data.The data set contains 12 aspects.By dealing with the missing value and abnormal value of these variables,and using woe box to screen out the effective variables for the experiment.2.Improve the efficiency of logistic regression model.In order to solve the problem that the logistic regression model can not select features,a composite model combined with gradient descent tree is proposed to improve the performance of the logistic regression model.3.Improve lightgbm algorithm.By modifying the value of lgbparameters,the experiment shows that the prediction ability of lightgbm model is improved.4.Optimize the scoring model of applicants.The xgboost model and the improved lightgbm model are used as the first layer basic learners of the hierarchical integration framework to learn,and the output results of the basic learners are taken as features and then brought into the second layer for training.The experimental results show that the prediction ability of the improved applicant scoring model is 6% higher than that of the logistic regression model.