| Credit card business is an integral part of the overall strategic layout of the bank,and it is also the advantage of the bank to stand firm in the industry.With the continuous opening of the financial market and the renewal of the financial industry,the influx of all kinds of financial products is bound to bring huge competitive pressure to the credit card business.For the credit card business,stock customers will bring sustainable economic benefits,which is the "engine" of the bank’s sustainable development.Therefore,predicting the potential loss of customers,retaining existing customers,strengthening customer relationship management and improving customer loyalty will become one of the core work of bank development.Catboost algorithm is a novel algorithm proposed in 2017,which is a further improvement on gbdt and xgboost methods.Its advantage is that it can automatically process category variables,and adopts sorting promotion method to overcome the prediction bias problem caused by traditional gradient bias.At present,the research on credit card customer flow at home and abroad mainly uses some mature machine learning methods,such as random forest,xgboost,stacking and so on.This paper analyzes the loss of credit card customers,according to the competition data of credit card customers in kaggle platform,carries out data preprocessing,feature engineering and other work,and uses catboost algorithm to establish a two classification model.Finally,according to the evaluation index of the model,it compares with other models,and further analyzes the interpretability of the model by using the SHAP value to verify the effectiveness of catboost algorithm Effectiveness and superiority.The practical results show that the prediction results of catboost algorithm are similar to those of xgboost algorithm,but the prediction accuracy has been further improved. |
