The Research Of Data Mining In Telecoms Customer Retention

The telecommunication enterprises gradually became aware of a customer- centric business need and started to change business type from business-driven to customer-driven. development of telecommunications technology and innovation have produced a constant variety of new services, and resulted in that a large number of low-loyalty customers switch to other networks or change the business. How to maintain customers, to extend the customer life cycle? This paper examined from three aspects: (1)Subdivide the market and customers; (2)Sell suitable business to the most needed customers and to best match the business and customer; (3)Effectively curb the phenomenon of large-scale customer attrition.In this paper, to address these issues, we studied a number of data mining techniques and methods. We purposefully analyze customer subdivision, cross-selling and update-selling and customer attrition analysis of the telecommunications industry. And integrate the improved mining algorithms and proposed ideas into them.Aiming at customer subdivision, we proposed a new clustering algorithm that is PSO-based K-means clustering algorithm. The theoretical analysis and experimental results show that the algorithm has better global convergence, and can effectively overcome the shortcomings of traditional K-means algorithm which is that it is too easy to fall into local minimum. Then we use the improved algorithm to analyze customer consumption data of one city telecommunications, and use cluster analysis, then divide them into customer groups with different characteristic features according to telephone fare. Experiments show that the analysis of clustering results more reasonable clearness and more convenient for different groups to adopt different business strategy,and provide managers with a reasonable decision support.We use Apriori algorithm which is the most classic association rule mining algorithm in analyzing cross-selling. But there are two shortcomings: first is too many times of the massive database scan, second is using the connect operation to generate candidate frequent item sets. We proposed an improved method mainly directed at these two shortcomings, it not only saves the system spending, but also reduces the number of redundant candidate sets generated, and greatly improved the efficiency of the algorithm. Experimental data proved the improved Apriori algorithm is more feasible.In this paper we analyzed the classification of customer attrition, attrition ratio and the main reason. We put forward a customer attrition analysis methods using C5.0 Decision Tree as the primary means.Another focus of this paper is the evaluation of mining algorithms, mainly about the PSO-based K-means clustering algorithm and assessment of the classification model.