ABSTRACT Competition in the mobile telecommunications industry is becoming more and more fierce. In order to improve mobile operator's competitiveness and customer value, several data mining technologies can be used. One of the most important data mining technologies is customer clustering analysis. This targeting practice has been proven manageable and effective for mobile telecommunications industry. Most telecommunications carriers cluster their mobile customers by billing system data. This paper discusses how to cluster mobile customers based on their call detail records and analyze their consumer behaviors. Finally, an application of a mobile customer clustering analysis is given in this paper. INTRODUCTION Competition in the mobile telecommunications industry is becoming more and more fierce. Mobile operator's profits and ARPU (Average Revenue Per User) are facing tremendous challenges. Customer's demand become diversified, differentiation and requirements of service quality become more rational and strict. In order to improve mobile operator's competitiveness and customer value, several data mining technologies can be used. One of most important data mining technologies is customer clustering analysis. The aim of clustering is to categorize prospective customers into distinct groups for distinctive contact strategies and proximal offerings (Adriaans & Zantinge, 1996; Berson & Smith, 1997; Mattison, 1997; Russell, 1996; Berry & Linoff, 1997; Russell & Lodwick, 1999). This targeting practice has been proven manageable and effective for mobile telecommunications industry. Most telecommunications carriers cluster their mobile customers by billing system data. Billing system data describe customer subscribe, spend and payment behavior. Call detail records describe customer utilization behavior. They have more information to describe customer behavior than billing system data. Therefore, this paper discusses how to cluster mobile customers based on their call detail records and analyze their consumer behaviors. Finally, an application of a mobile customer clustering analysis is given in this paper. K-MEANS CLUSTER METHOD There are many clustering method, for example, fuzzy clustering method, system clustering method, dynamic clustering method and K-means clustering method. But the K-means method of cluster detection is most commonly used in practice. It has many variations, but the basic form is unchanged. Suppose that the number of mobile customers is P. Each mobile customer is described by an n-element vector Xi= ([x.sub.i1], [x.sub.i2], ..., [x.sub.in],), i = 1, 2, ..., P. K-means clustering method can be briefly stated as follows: 1) Select K data points or K mobile customers to be the seeds. The seeds are described by n-element vectors j = E([e.sub.j1], [e.sub.j2], ..., [e.sub.jn]), j = 1, 2, ..., K. MacQueen's algorithm simply takes the first K mobile customers records. That is, let [E.sub.j] = ([e.sub.j1], [e.sub.j2], ..., [e.sub.jn],) = [X.sub.j] = ([X.sub.j1], [X.sub.j2], ..., [X.sub.jn]), j = 1, 2, ..., K. 2) Calculate the distance of between each seed and every mobile customer by the following formula: [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Here.q is a constant. Usually, let q = 2. 3) Every data point or mobile customer is assigned to one seed according the principle of minimum distance Calculated by step 2). In this way, we get P group data sets or P group mobile customers. 4) Calculate the centroids of P group data sets and form new K seeds by the following formula: [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] Here, mj denotes the number of j group mobile customers. 5) Return to step 2) or stop if the changes of cluster boundaries are small enough. It is not difficult to program K-means clustering method mentioned above by software engineering. …