| With the rapid development of information technology,the telecom market tends to saturation,the telecom industry is facing fierce competition,serious homogeneity,customer loss and many other challenges.Among these messages,one of the most urgent problems that operators need to deal with is the loss of users.The so-called "loss of users" refers to the loss of interest in a product or service that users have never used because of the bad experience.But for a company,the cost of acquiring a new user is much greater than the cost of maintaining an old user.Therefore,if we can make use of the huge amount of user data to accurately predict the trend of user loss and retain the potential user loss,we can make the company’s overall revenue increase.Based on the user data of a certain telecom operator,this paper uses the fusion technology of Light GBM tree model and logistic regression model to forecast the trend of user churn.The purpose of this paper is to explore the model fusion method to compare the application of the traditional single algorithm in the early warning model of user churn,and to assist the telecom operator to make scientific and reasonable decision.Firstly,this paper analyzes the importance of building a more accurate early warning model for the loss of telecom users,expounds the value and prospect of the model fusion technology.Then,the paper describes the model of customer churn in detail,including logistic regression model,decision tree model,stochastic forest model and Ligh GBM model.Then it introduces the characteristics of Light GBM +LR fusion model: LR linear model is simple,efficient,but does not consider the relevance between features.Light GBM tree structure can achieve automatic fusion and feature selection of multidimensional data.Secondly,we preprocess the collected data sets,including missing value filling,data type conversion and so on.Then,the paper makes a descriptive statistics and exploratory analysis of the processed data,so as to have a clear understanding of the relationship between the characteristics of telecom customers and the behavior of customer churn.Finally,a prediction model of the loss of telecom operators is constructed to predict the potential loss of users.Firstly,the parameters of each model are set and explained.Then,compared the model effect before and after feature derivation,the AUC value is 0.022 higher than that of the LR model without feature creation,and the fusion model based on artificial field construction + tree model predicts the AUC value to 0.989,which proves the effectiveness of the method.Furthermore,the single model and Light GBM + LR fusion model are compared and evaluated,and the results show that the AUC and AP values of the fusion model are significantly higher than those of single model.According to the analysis,the key factors that affect the loss of customers are obtained,and feasible suggestions are provided for telecom operators to reduce the loss of customers,so as to obtain more income and increase profits. |
