Research On Feature Selection Method In Bank Credit Card Default Prediction

As credit cards bring convenience to people,more and more people use credit cards.In order to seize the growing market,a lot of banks are trying to acquire as many customers as possible,which results in the easy issue of credit cards to unqualified users and more default payments.Heavy interests loss of banks makes it an urgent job to control the default payments.The credit card default prediction,the early discovery of potential default repayment users,can provide an important basis for default control so that banks can take effective measures in time.With the rapid development of information technology and the gradual improvement of the bank's data system,the consumption data of credit card users is becoming more and more abundant,and its scale and complexity are also growing.By digging deeper into and analyzing credit card consumption data and exploring hidden patterns in consumer behavior,banks can not only further understand the consumer behavior of credit card users,but also select useful information for default prediction.Therefore,in response to the credit card default prediction problem,this paper studies the feature selection and proposes two feature selection methods based on indepth exploration and analysis of the credit card consumption data.The main work of this paper is as follows:(1)Excavating and analyzing a bank's credit card consumption data set from the perspective of the network,an interesting pattern is fount between features,namely: the feature clustering phenomenon.This pattern not only reveals the complex relationships between features,but also helps to discover features with a large amount of information.(2)Based on the revealed feature patterns,a community-based feature selection method is proposed,and then its feature selection results are used to train credit card default prediction models.The experimental results from real data show that the proposed community-based feature selection method has better prediction effect in the gradient decision tree model and K-nearest neighbor model,and the prediction effect is more prominent in the random forest model.(3)In order to further improve the default prediction,a community-based and reweighted feature selection method is proposed.Firstly,each feature is weighted according to the correlation between the feature and the tag.Secondly,the feature selection is combined with the feature pattern and the feature weight.Finally,the selected feature is used as an input training credit card default prediction model.The experimental results from real data show that the proposed community-based and reweighted feature selection method has improved accuracy,recall,precision,and F1 score on the Na?ve Bayesian model,the Logistic regression model,and the K-nearest neighbor model.The results of some evaluation indicators in other models have also improved.At the same time,c community-based and re-weighted feature selection method combined with random forests provide the most accurate prediction of credit card defaults.