Churn prediction in the telecommunication industry using social network analytics

Identifying potential churners is important in competitive and saturated markets such as the telecommunication industry. Research has shown, that performance of models that predict customer churn is increased when social network methods are applied. Studies also indicate that relational learners, applied to customer call networks, are capable of predicting churn accurately. In this research, the difference in performance of a variety of relational learners is tested by applying them to a number of CDR datasets from around the world. In addition, performance of relational classifiers and collective inference methods is compared and the performance of models which combine relational learners with mainstream classifying methods, such as logistic regression, is studied. Our results show that collective inference methods do not improve the performance of relational classifiers and that the scores of relational learners combined with other customer features result in the best performing models.status: publishe