Prediction Of High Value Customer Churn In Commercial Banks Based On Machine Learning

Customers are the fundamental interests of commercial banks.With the accelerating process of information construction,the competition among banks for customers is becoming increasingly fierce.At the same time,internet finance also diverts customers with its advantages of low threshold and high return.The competition between banks has turned into competition for customers.Who can have a group of stable customers,who can get the initiative of market competition.Generally speaking,20%of high-quality customers contribute more than 80%of profits.So,how to predict high-value customers with potential loss risk in advance is the major focus of commercial banks.Machine learning can simulate human learning behavior and extract valuable information from data.Based on the real demand of a commercial bank,and with the help of machine learning method,this paper identifies and mines high-value customers for the bank,and establishes a customer churn prediction model.The main work of this paper includes:First,conduct customer portrait on all existing customers of commercial banks and identify high-value customers.Use K-means cluster analysis method to comprehensively classify customers from multiple dimensions of asset value,loyalty and activity.Customers with diamond asset value,high loyalty and high activity are defined as high-value customers and labeled.Second,taking high-value customers as the basic customer group,combined with business logic and understanding,repeatedly analyze the massive customer information and behavior data to determine the specific definition of customer churn.The loss of high-value customers is defined as the continuous conversion of high-value customers in this month to non high-value customers in the next three months.After comparing the advantages and disadvantages of a variety of supervised learning algorithms,the algorithm of combining decision tree and logistic regression is selected finally which it is strong in explanation and easy to generate score card system.The application effect of the model is evaluated by KS value,AUC value and Gini coefficient.It is consistent that the model has good performance and strong discrimination ability.Third,in order to facilitate the understanding and use of business personnel,the probability value of the model output result is transformed into the corresponding specific score between 350-950 through the scoring system.The establishment of the model helps commercial banks identify existing customers,screen out high-value customers,and predict customers with potential loss in advance,so that customer managers can carry out customer management and maintenance.It not only saves a lot of labor costs for commercial banks,but also brings potential economic benefits and reduces the losses of commercial banks.