Measuring the Importance of Churn Predictors in Romanian Telecommunication Industry

Abstract

Telecommunication sector is a saturated market, and each client’s action affects the company profit. The most valuable asset of the telecommunication company represents its clients’ database. The Telecom industry pays special attention to the migrant clients because from an economic perspective, the cost invested by the company in the acquisition of a new customer is higher than the cost of keeping an existing client. This paper aims to determine the most important factors that influence the decision of a client to migrate from a telecom provider to another through a graphical method. We apply the churn prediction model on a dataset from Romania that has not yet been studied before. We choose to use the Balanced Random Forest technique to build the churn prediction model and the AUC coefficient to evaluate it. Permutation importance makes a classification of the most important features in the model and measures their impact through a metric called the importance score. The result proves that the most significant three factors in the churn phenomenon (client migration) are the number of months since the last change in the account, the number of minutes off the network and the invoice cost, a significant difference in score, the first the indicator being ten times more important than the next one. Therefore, we can state that we can resolve the main action by resolving the churn problem on the current dataset solved by monitoring and evaluating these variables.