The K-means Algorithm Improvement And Its Application In The Customer Segmentation Of The Communications Industry

With the development of communication services, the business model of thedomestic communication companies based on products has gradually been shifting tothe international advanced mode based on customer data and information. Therefore,customer segmentation has become the prerequisite and foundation of this mode. Howto reasonably segment the consumers based on their behavior, facilitating to provideevery customer with specific services, aiding to create customer value and the sametime pursuing enterprise’s profitable maximization have become key issues forcommunications operators. Data mining clustering technology helps to clustercustomers which have the same characteristics into one class, so as to providetechnical support for customer segmentation.The k-means algorithm of clustering technology and application of k-meansalgorithm in customer segmentation of communication industry are studied in thisthesis. Major works are as follows:Firstly, the k-means algorithm of the clustering analysis is studied in this thesis.The k-means algorithm ideas and basic processes are described. Advantages anddisadvantages of the algorithm and the existing improvements are analyzed. It focuseson the dependence of the k-means algorithm to the initial clustering center and the timeefficiency of the k-means algorithm that still needs to be improved when dealing withhuge amounts of data.Secondly, as to these two shortcomings, an improved algorithm is proposed in thisthesis. As much as possible to stay further apart the sample points are taken as initialcluster centers in the improved algorithm. It can avoid randomly selected centersgetting too close in the classical algorithm. Consequently, it can effectively prevent theobjective criterion function into a local optimum. On the other hand, the triangleinequality principle is used to improve, when calculating and comparing the distancebetween sample points. This method avoids unnecessary comparison and distancecalculation and increaseing the algorithm’s time efficiency. Then, the simulationexperiments demonstrate that the improved algorithm is superior to classical k-meansalgorithm in clustering quality, stability and the time efficiency.Finally, the improved algorithm is applied to the telecom customer segmentationin this thesis. We get ideal result and analyze this result. So marketing strategies areset up to different customer groups and reasonable decision supports are provided for the enterprise. Thus, the subject has a certain practical significance and applicationvalue.