3 D. McGowan A. Brew B. Casey N.J. Hurley 2011 Proceedings of the 22nd Irish Conference on Artificial Intelligence and Cognitive Science In telecommunications network analytics, a problem of significant interest to service providers is churn prediction, that is, the identification of customer’s who have a high probability of leaving the network in the near future. From a business perspective, it is understood that it is more beneficial financially to retain a customer rather than acquiring a new one. It is understood that churn exhibits a viral component, in the sense that if one customer churns, friends in his immediate social network are likely to also churn. This paper presents an evaluation of supervised machine learning classification algorithm’s applied to the problem of churn prediction. A number of classifiers including regression, boosting and dimension reduction techniques are examined. We find that generalised linear models perform the best and that feature selection can provide better performance on a reduced feature set, thereby increasing the efficiency and accuracy of the classification.