Dyadchurn: Customer churn prediction using strong social ties

The increase in mobile phone subscriptions in recent years, has led to near market saturation in the telecom industry. As a result, it has become harder for telecom providers to acquire new customers, and the need for retaining existing ones has become of paramount importance. Because of fierce competition between different telecom providers and because the ease of which customers can move from one provider to another, all telecom service providers suffer from customer churn. In this paper, we propose a dyadic based churn prediction model, DyadChurn, where customer churn is modeled through social influence that propagates in the telecom network over strong social ties. We propose a novel method for evaluating social tie strength between telecom customers. We then, incorporate strong social ties in an influence propagation model to predict the set of future potential churners. The evaluation of the proposed dyadic based churn prediction model has been done using a real dataset, from one of the largest telecom companies in Egypt. The experimental results showed that the "length of calls" between customers is the most effective attribute in predicting social influence that result in churning. The results also showed that strong social ties (as opposed to weak ties) were the most effective ties in determining churn. Using strong social ties only enhanced the prediction accuracy (in terms of the lift curve) by more than 20%, when compared to a diffusion model. © 2017 Association for Computing Machinery.