Credit Card Default Prediction Based On Weighted Stacking Ensemble Learning

In recent years,the credit card business has developed rapidly and has become an important profit growth point for China's banking industry.No mortgage is the biggest feature of credit card,which brings a great convenience to customers.China's credit card business started late,and there are problems such as imperfect system and irregular process in the development.Therefore,while credit cards bring high returns to banks,they also bring certain financial risks.How to effectively use the historical data of credit card customers to avoid risks is a topic of concern to many scholars at home and abroad,and it is also what banks should focus on.In the experiment of this paper,a variety of machine learning algorithms are used to establish a classification model that can effectively identify default customers of credit card based on actual data,further discovering the potential default risk of credit card customers in advance,and reduce the losses caused by credit card defaults to banks.Based on the existing knowledge and actual situation,this paper analyzes the credit card business in China and introduces the basic theory of the machine learning algorithm selected in the experiment.Then,taking the credit card business data of Taiwan as an example,the corresponding data processing is performed.After that,with Python programming,a single data mining model for credit card default risk identification such as decision tree,support vector machine,Random Forest,Xgboost and LightGBM is established for the data,which is finished processing,and indicators such as overall classification accuracy,default recall rate,₁ score and AUC,which are selected from the model,are used as criteria for evaluating the classification effect of the model.Finally,three machine learning algorithms of Random Forest,Xgboost and LightGBM are selected as the primary learners in Stacking ensemble learning,while logistic regression are chosen in the secondary learners in the second phase.At the same time,based on the classification error rate of the default class of the primary learner,the primary learner is weighted,and the default recognition combination model of weighted Stacking ensemble learning is established.The final research results of this paper show that it can effectively predict credit card default customers and have more accurate classification effect than the single classification model mentioned in the paper as long as weighted Stacking ensemble learning strategy and logistic regression are used as secondary learners based on historical data of credit card customers.Through the predicted classification results of the test,the model can identify more default customers.The final test results show that using the weighted Stacking ensemble learning of classification model can improve the recall rate of the defaults to a certain extent,and the comprehensive performance of the classified model is also optimal.Therefore,the research results of this paper can give the bank a certain degree of help with the management of credit card.The classification forecast of credit card customers can identify the potential default risk of customers,who apply for credit card,and strengthen the bank's management of credit cards and reduce the default rate of credit cards.