Market Segmentation of Mobile Internet Customers Using Clustering Algorithms: The Case of Ethio Telecom

Telecom companies utilize data mining algorithms and tools to understand the behaviors of their customers. Cluster analysis is one of the techniques used to identify homogenous groups of customers from a heterogeneous group based on the customers' service usage records. Clustering algorithms aim to find natural groupings of subscribers and are widely applied for customer profiling and market segmentation. As telecom customers use services like cellular Internet, huge amount of data is generated in the form of call detail record (CDR) primarily for billing purpose. This data can also be a source of information for marketing and network management tasks. Previously, different algorithms have been studied and implemented by researchers to understand how, why, when and where people access mobile Internet using already available data from the telecom systems. In this thesis, two clustering algorithms namely K-means and Two-Step were implemented on real CDR and CRM datasets of a telecom company in Ethiopia to segment mobile Internet users. Behavioral segmentation was performed using aggregated data of sample number of customers based on derived features. Moreover, the insights obtained from each segment were analyzed before suggesting marketing strategies for personalized services and targeted campaigns. K-means was applied on CDR dataset and meaningful clusters showing characteristics of users were obtained and explained. Two-Step clustering was found to be more suitable for segmenting user groups and has a better silhouette score than K-means. The results of the analysis show the possibility of using data stored by mobile operators for market segmentation purposes. It was examined from this research that clustering algorithms like K-means and TwoStep are applicable to observe patterns among ethio telecom customers based on service usage and demographic information of subscribers from database systems .