Churn Analysis in a Romanian Telecommunications Company

Telecommunications is one of the sectors where the customer base plays a significant role in maintaining stable revenues, so special attention is paid to prevent their migration to other providers. Over time, businesses in the telecommunications industry have faced multiple threats of financial loss from migrating customers who want to leave their telecom service provider in exchange for other offers from competing companies. An effective prediction model of this action can not only be viewed as an insurance policy, supporting stable revenue, but also provides suggestions for database management so that potential migrant customers can benefit from personalized offers and services, depending on their profile, thus preventing their loss. The aim of this paper is to predict customers who are ping to defect in a Romanian mobile telecommunications company. The chum analysis is developed for post-paid customers. We used logistic regression to predict chum and a solution based on smoothed bootstrap technique to correct for the drawbacks of imbalanced classes. In our study this procedure did not significantly improve the performance of the logistic classifier measured by AUC (Area Under the Receiver Operating Characteristic curve). So even after balancing the sample we still obtain a really reduced value of the AUC, making it difficult to correctly predict chum phenomenon on the available data set.