Customer Churn Prediction in Telecommunications

This study is aimed at customer churn prediction in telecommunication company Er-Telecom Holding. Due to the characteristics of the data of various companies, there is no single method for churn prediction. In this regard, to solve the problem, two machine learning algorithms were chosen - logistic regression and gradient boosting. These methods are needed in order to choose the most relevant for the company's goals. The main criterion for choosing the best algorithm is recall - a criterion that considers a 1 type error, that is, assignment of churn customers to non-churn ones. To build models, one customer segment was chosen - loyal customers with the Internet. The results showed that even though logistic regression is a basic method for solving classification problems, it is significantly inferior to the most advanced method - gradient boosting. Thus, to solve the customer churn problem, it is necessary to consider several machine learning algorithms in order to choose the best one. For the segment “loyal customers with Internet” it is recommended to use gradient boosting.