Застосування логістичної регресії для прогнозування відтоку клієнтів телекомунікаційної компанії

Abstract

This work is devoted to solving of customer churn prediction problem. The mathematical model of this problem is logistic regression, where inputs are cus-tomers characteristics and output – prediction if customer is going to refuse company’s services. To determine major features the following methods are used: recursive feature elimination, univariate feature selection with chi-square test and mutual information. As a result, a linear classifier was built and esti-mated on real data.