Customer Segmentation and Customer Profiling for a Mobile Telecommunications Company Based on Usage Behavior

Vodafone, an International mobile telecommunications company, has accumulated vast amounts of data on consumer mobile phone behavior in a data warehouse. The magnitude of this data is so huge that manual analysis of data is not feasible. However, this data holds valuable information that can be applied for operational and strategical purposes. Therefore, in order to extract such information from this data, automatic analysis is essential, by means of advanced data mining techniques. These data mining techniques search and analyze the data in order to find implicit and useful information, without direct knowledge of human experts. This research will address the question how to perform customer segmentation and customer profiling with data mining techniques. In our context, ’customer segmentation’ is a term used to describe the process of dividing customers into homogeneous groups on the basis of shared or common attributes (habits, tastes, etc). ’Customer profiling’ is describing customers by their attributes, such as age, gender, income and lifestyles. Having these two components, managers can decide which marketing actions to take for each segment. In this research, the customer segmentation is based on usage call behavior, i.e. the behavior of a customer measured in the amounts of incoming or outgoing communication of whichever form. This thesis describes the process of selecting and preparing the accurate data from the data warehouse, in order to perform customer segmentation and to profile the customer. A number of advanced and state-of-the-art clustering algorithms are modified and applied for creating customer segments. An optimality criterion is constructed in order to measure their performance. The best i.e. most optimal in the sense of the optimality criterion clustering technique will be used to perform customer segmentation. Each segment will be described and analyzed. Customer profiling can be accomplished with information from the data warehouse, such as age, gender and residential area information. Finally, with a recent data mining technique, called Support Vector Machines, the segment of a customer will be estimated based on the customers profile. Different kernel functions with different parameters will be examined and analyzed. The customer segmentation will lead to two solutions. One solution with four segments and one solution with six segments. With the Support Vector Machine approach it is possible in 80.3% of the cases to classify the segment of a customer based on its profile for the situation with four segments. With six segments, a correct classification of 78.5% is obtained.