Telecom Customer Churn Prediction Based on Classification Algorithm

Abstract

All walks of life are generally concerned about the problem of customer churn. This paper deals with the real telecom customer characteristic data and constructs a prediction model based on classification algorithm to predict the customer churn rate. Among the known classification algorithms, XGB algorithm performs better, its accuracy is 79.98%, recall rate is 90.17%, F1 is 84.21%. According to the study, users with the following characteristics are more likely to churn (in descending order of churn rate) : tenure 1-5, use "e-check", TotalCharges is less than ࿥2,281.92, MonthlyCharges is more than ࿥64.76 , with electronic bills.