The Prediction of Subscribers' Churn in the Mobile Telecommunication Company

The work raises the problem of customer churn in the telecommunications sector. At the moment, the telecom market is oversaturated and operators are paying increasing attention to retaining the existing subscriber base than on increasing the number of new sales. This work aims not only to predict the future outflow of customers by building a predictive outflow model, but also to compile a matrix of special offers for customers who are prone to outflow. The construction of such a matrix is ​​based on a study of the causes of customer churn and comparison of these causes with the customer profile of using mobile communications. The study found that the most often customers refuse services of one provider in favor of another because of problems with network quality or too high subscription fees for the tariff. The predictive outflow model in the work is built using the random forest method and allows you to select micro-segments of customers who are ready to leave in the near future. The model was based on historical data on the use of the network and some other data on the client for 30 days, the predictor of outflow was inactivity (absence of any transactions) in the next period of more than 7 days. It is also worth noting that the sample was formed from the subscribers of the St. Petersburg branch of a Russian telecom operator and included subscribers with a network lifetime of more than three months. As a result of the study, an attempt was made to compile a matrix of offers relevant to different customer segments for their retention. The financial effect from the implementation of such initiatives should be positive.