Research On Credit Card User Loyalty Forecast Based On Data Mining

In recent years,the rapid development of Internet finance has caused a major impact on the credit card business of commercial banks.The credit card business has formed a major source of profit for retail banking.How to maintain customer loyalty to the bank has far-reaching significance for the credit card business that is currently invaded by Internet finance.At present,the empirical research on customer loyalty in academia mainly studies the influence variables of customer loyalty in the context of specific industries;the research on customer loyalty prediction mainly focuses on the exploration of different loyalty models.However,the data used is basically questionnaire data or the time span of the data is small(for example,one month),and there are fewer studies on data sets with larger time spans.In view of the above,this paper focuses on mining effective indicators from credit card customer transaction data with a large time span,and groundbreakingly proposes a credit card customer loyalty prediction model based on the XGBoost algorithm.Specific work includes the following:1.The customer's real credit card transaction record data was subjected to data preprocessing such as attribute screening,data type conversion,and missing value processing.Due to the large time span of the data used in this paper,the data preprocessed indicators are rough,which is not conducive to customers The prediction of loyalty,therefore,based on the evaluation index system of the objective attribute data of customers,the indicators of customer loyalty influencing factors were subdivided.2.By analyzing the importance of features,it is found that the number of purchases,whether it is a gold card,and the maximum consumption amount are the three most important factors affecting credit card customer loyalty.In addition,in the top half of the characteristics,four characteristics belong to the characteristics of customer behavior,and belong to the attributes generated by subdividing the indicators.This shows that when modeling credit card transaction data with a large time span in this paper,feature construction is very necessary.3.Contrast research on credit card customer loyalty prediction.The four basic models of XGBoost,logistic regression,SVM,and decision tree are used to build a customer loyalty prediction model framework.In this paper,a comprehensive analysis of the results of the two loyalty customer discrimination accuracy rates(accuracy rate)and the entire sample discrimination accuracy rate is used to obtain the conclusion that XGBoost has the best prediction performance.