| Customer churn is a common problem in the banking industry in today’s increasingly competitive commercial banks,and it has a significant impact on the banks’ finances.The financial technology(Fin Tech)development plan(2019-2021)issued by the People’s Bank of China proposes that "on the premise of effectively protecting personal privacy,commercial secrets and sensitive data,establish a sound mechanism for data integration and application,and realize effective integration and in-depth utilization of data resources".As a serviceoriented financial institution,customer deposits are an important source of funds for banks.Reducing customer churn can help banks improve customer retention,increase profits,and gain competitive advantage in the market.As a financial institution with a large amount of customer data,banks are able to use the data to make better realizations for their own development.Studying deposit customer churn is a complex problem that requires sophisticated analytical tools and techniques to solve.The research of this paper aims to explore the development status and shortcomings of the customer churn prediction model of commercial banks.To this end,this article will start with the research background and the research status at home and abroad,and conduct in-depth research and analysis on the frontier work in this field.Then,based on the previous work,the concept and principle of the customer churn prediction model in this paper were proposed,and the comprehensive evaluation index method of ROC curve,AUC value and F1 score was used to carry out the model using a data set from a commercial bank’s real historical data.verification.In terms of feature screening and feature expansion,this paper uses the method of comprehensive scoring screening to select strong features,and designs a set of feature expansion methods based on natural logarithms to generate new data sets.Finally,we designed an ensemble learning model based on Boosting,conducted experimental analysis on the newly generated dataset,and verified the performance and superiority of the model by means of experimental analysis and comparative analysis.The research results show that the feature screening and feature expansion methods designed in this paper can increase the training efficiency and accuracy of the model,and the constructed model is better than the current mainstream machine learning single learning method in terms of predictive performance for new data sets.It can better help banks understand their customers,formulate more effective strategies to retain customers,and provide reference for other industries to solve customer churn prediction problems.The research results of this paper will provide a basis for the development of commercial bank customer churn prediction models.The new ideas and methods have certain theoretical and practical significance. |
