Please use this identifier to cite or link to this item: http://hdl.handle.net/1946/10159
In 2010, the penetration of the Icelandic mobile telephony market has reached about 120%. Competition is fierce in such a highly saturated market. Customers become more and more demanding on price and service. New regulations and technologies allow them to switch easily between mobile operators. As the result, customer churn has increased significantly. Facing this challenge, mobile operators shift their attention from customer acquisition to customer retention. The crucial elements of customer retention are accurate churn prediction models and effective churn prevention strategies. The goal of this study is to construct a churn prediction model that can output the probabilities that customers will churn in the near future. Churn prediction is formulated as a classification task of churners and non-churners. Learning algorithms are applied on training data to build classifiers. The data is a set of customers where each one is represented by numerous features and labeled as churner or non-churner. The primarily step involves employing feature selection to search for relevant features, eliminate irrelevant or redundant ones. Afterwards, the reduced data with only relevant features are passed into classifiers trained using machine learning algorithms: (1) C4.5 decision tree, (2) alternating decision tree, (3) Naïve Bayes and (4) logistic regression. The result of this study is twofold. Firstly, churn indicators are identified and insights are provided into churn behavior in postpaid and prepaid sectors for the time period from July 2010 to May 2011. Secondly, churn can be forecasted with a certain accuracy in advance which enables the mobile operator to carry out suitable reactions.
Keywords:mobile telephony market, churn prediction, feature selection, machine learning, datamining, Siminn
|Customer Churn Prediction for the Icelandic Mobile Telephony Market.pdf||3.39 MB||Open||Heildartexti||View/Open|