Customer Churn Prediction Based on Machine Learning

Abstract

In recent years, telecommunication has become the major media of communication, while the market has been saturated in many developed countries. Telecommunication companies are shifting their concerns from acquiring new customers to retaining regular customers, since the cost to win a new customer is higher than that of maintaining a regular customer. Therefore, it is essential for telecommunication operators to forecast customer churn in this competitive, accurate market. This paper aims to identify affecting customer churn and construct an efficient model, which is used to predict and analyze data with visualization results. The churn forecast consists of several phases: data prepossessing, data analysis, evaluation measure, and application of machine learning algorithms. Moreover, data pre-processing covers data cleaning, transformation, and classification. The machine learning classifiers selected are Logistic Regression, SVM, Random Forest, AdaBoost, GBDT, XGBoost, Light GBM, and CatBoost. Classifiers were evaluated using performance measures, such as accuracy, precision, recall, AUC, and F1-Score. Based on the paper, the result was shown that the Light GBM outperformed other classifiers while identifying potential churners.