Approach to Prediction of Mobile Operators Subscribers Churn

Abstract

This article presents an approach to the description of machine learning methods for predicting the outflow of telecom operator subscribers. Describes the parameters characterizing the interaction of the mobile operator with end users. The parameters that have the greatest influence on the client’s decision to refuse the services of a mobile operator have been determined. The originality of the approach lies in the use of such mathematical methods that allow you to determine the main set of parameters, due to which specific subscribers are prone to changing the mobile operator. The proposed approach allows you to organize a system in which it is possible to determine the main parameters characterizing the tendency of customers to outflow and acting on them using various methods to try to increase subscriber loyalty. A comparative analysis of the results obtained using the analyzed logistic regression methods, Bootstrap aggregating and random forest showed that the spread of prediction errors does not exceed 6%. However, the advantage of the random forest method is the ability to determine the set of parameters that make the greatest contribution to making decisions by the subscriber to change the mobile operator. Therefore, for analyzing customer loyalty, a random forest method can be recommend, which showed on the test sample an improvement in the accuracy of the predictions in the sample to 6-7%.