PREDICTION OF THE OUTFLOW OF MOBILE OPERATOR SUBSCRIBERS BASED ON MACHINE LEARNING METHODS

Abstract

The research article introduces an approach to describing machine learning methods for predicting customer churn in a telecommunications operator. It outlines parameters characterizing the interaction between the mobile operator and end-users, identifying those with the most significant influence on customer decisions to churn. The novelty of the approach lies in utilizing mathematical methods to pinpoint the primary set of parameters driving specific subscribers to switch mobile operators. This proposed approach enables the organization of a system to identify key parameters indicating customer churn tendencies and intervene with various methods to enhance subscriber loyalty. A comparative analysis of results obtained using logistic regression, initial load aggregation, and random forest methods showed that the prediction error rate does not exceed 6%. However, the advantage of the random forest method lies in its ability to identify the parameter set contributing the most to a subscriber's decision to switch mobile operators. Thus, for analyzing customer loyalty, the random forest method is recommended, demonstrating an increase in prediction accuracy by 6-7% in the test dataset.