Carrier Customer Churn Early Warning Model Based On Logical Decision

In the modern Internet industry,the behavior of user networks generated and accumulated by Internet companies in their daily operations is extremely large.By constructing a loss warning model,the characteristics of the lost customers are analyzed in advance to identify the customers that are about to be lost,and to retain them.Relying on retaining stock customers is the top priority of business management.In the early warning of customer churn,two main considerations are:first,predicting the overall performance of the model;second,identifying important drivers of customer churn and analyzing their causes of churn.Therefore,the churn prediction model should have good predictive performance while being easy to implement and easier to understand.Decision trees and logistic regression are two very popular algorithms in customer churn prediction,with strong predictive performance and good comprehensibility.Despite these advantages,decision trees have some shortcomings in dealing with the linear relationship between variables,and logistic regression has certain disadvantages in terms of the interaction between variables.Therefore,for these two problems,this paper proposes to use a new hybrid algorithm,namely the logical decision model.The main difference between the two is that the algorithm is a different model built on the data segment instead of the entire data set.Not only can predictive performance be improved,but the models built in each blade are also easier to understand to better classify the data.This paper studies the commonly used algorithm model for customer churn early warning.Based on the public data set Churn as the basic case,after preprocessing the data,the decision tree and logistic regression are used to predict and analyze,and the two methods are combined.The model is constructed using the new hybrid modellogic decision algorithm.The accuracy of the tested model is improved from 0.76 to0.891,the AUC is from 0.812 to 0.912,and the F1-Score is from 0.456 to 0.686.The prediction accuracy of the logic decision model algorithm is higher.Mainly done the following work:(1)Before the model is constructed,the data is preprocessed,the SMOTE algorithm is used to deal with the imbalance of sample data,and the sample data is re-selected for training and prediction.(2)Based on the algorithm of logistic regression and decision tree,this paper adopts a logical decision model combining the two to include two phases: the segmentation phase and the prediction phase.In the first phase,the customer is subdivided according to the decision rules;in the second phase,an applicable logistic regression model is constructed for each leaf node of the tree.(3)In terms of predictive performance and comprehensibility,this paper uses AUC and F1 Score to evaluate the model.The experimental results show that the logical decision model score is significantly better than its building block logistic regression and decision tree.