Prediction of churn rate for prepaid customers in turkish telecom industry

Telecommunication sector is the one of the competitive markets in business world. It is evolved from telegraph to mobile services in a short span of time due to technologic developments. Therefore, competition between telecommunication companies has increased rapidly over time. As managing the competition among competitors, companies should know their customer behaviors and act preventively. In real world, it is hard to predict customers' behaviors and hence taking actions regarding each customer in companies' bases. Therefore, actions should be reduced and be focused on a specific group of customers. Also, it is known that new acquisition cost of the customer is much more higher than keeping the base (existing) customer not only in telecommunication but also in other industries. On the other hand, customer churn becomes easy with Mobile Number Portability (MNP) capability in the telecom industry. However, retention activities for churner customers are costly too. At this point, customer churn methods help companies to predict customers' brand switching behaviours to the other operators. Prediction of the churn provides detection of potential churners and helps to identify the strategies to keep them by the companies. Some algorithms and methods are offered in previous research for the calculation of churn possibility. In this thesis, it is aimed that, customer churn in telecom industry can be predicted with high degree accuracy using customer behavior dataset and decision tree method. This method is implemented on SPSS V22.0 program and the results are analyzed. In conclusion, behavioral and demographic dataset lead to churn rate calculation. Especially, offnet usage minutes which means call to other operators, out of bundle (OOB) usage minutes which means exceeded tariff minutes, customer tenure which means stayed months in customer base and customer age are the most accurate variables for predicting prepaid customer churn. The main limitation of the analysis lies in the issue that prepaid customers are churn flagged in the company system on the basis of a longer time period that is based on number of top-ups in specific months. The longer time leads to lower accuracy of potential churn customer since they were already gone. Because of that, decision tree models cannot help well for detecting the link between usages and churn rate properly in such circumstances.