Application of social network to improve effectiveness of classifiers in churn modelling

The subject of presented work is prediction of users willingness to churn - i.e. to change the provider of telecommunication services. The typical solution of this problem is application of classification methods to data inferred from the client call history. Classical methods of data mining are widely used by operators however their performance is often far from desired. The source of such a situation may be in neglecting or week modeling of social relations. The proposed approach consists of preparing standard regression model and augmenting it with data gathered by the construction and analysis of the social network. This way it is possible to exploit call history twice and build a model which is still easy to interpret.