TELECOM CUSTOMER SEGMMENTATION USING DATA MINING TECHNIQUES

Abstract

The aim of this research is to apply data mining techniques in telecom sectors to build models that can identify the contribution that customer makes to organization profitability based on current relationship with the organization. The objective of this research is to design enterprise customer segmentation model to Ethio Telecom that is used to identify the high value and behavior of enterprise customer. To meet the objective of the study we use hybrid data mining process model, which consists of six phases to undertake data mining process and to address the business problem. During the understanding of the problem, business practices of ET enterprise section are measured. This is done using interviews with business and technical expert and document analysis. Data preprocessing is done using different data mining methods. To prepare the data for analysis, we select 162315 records of customer data to conduct this research. After data preprocessing, we get 21126 records with thirteen attributes that are used for data mining task. This research is conducted using WEKA software version 3.8.2 and three clustering algorithm, namely, k-means, filtered and farthest first are used. Among clustering algorithms, farthest first clustering algorithm has better clustering performance than other cluster algorithms (filtered and k-means). Hence, the model constructed by farthest cluster is used to design a prototype. The result of this study is interesting and encouraging and confirmed that applying data mining techniques truly support customer segmentation activities at ET. In the future we recommended more segmentation studies by using a possible large amount of customer records and employing other clustering algorithm yield better results.