Vinsamlegast notið þetta auðkenni þegar þið vitnið til verksins eða tengið í það: http://hdl.handle.net/1946/31186
The Icelandic telecommunication industry is immensely competitive. The use of data mining and machine learning for churn prediction in the telecommunication industry has been researched to a great extent in recent years. In this paper, the application of these methods on the Icelandic telecommunication market is researched. CRISP-DM and machine learning methodology was used to create predictive churn models. CDR and customer service data was not used for the models. The results showed that predictive churn modeling gives promising results in the Icelandic telecommunication market. However to get satisfactory performance CDR and customer service data is likely needed. CRISP-DM seems to be an effective methodology for churn prediction. The models trained had similar ROC AUC performance but the best model to deploy in practice seems to be a neural network.