Fraud Detection On Telecommunication Business Base On Feature Tree Analysis

In this paper, we focus on the need for telecommunication business about the fraud problem, to guarantee the revenue issue. As we all know that telecommunication network is growing more and more complicated especially in China for catering for various classes of customers.The paper describes the antifraud system based on Data Mining theory and fraud feature tree construction. Data Ming provides an overall method framework in solving the problem while fraud feature tree construction is a detail on how the fraud detection is proceeding. Three elements come first before the feature tree is successfully constructed:telecommunication fraud characters analysis, subscriber data analysis, basic understanding on telecom business.The final solving method named'fraud verdict by nodes in feature tree'.Its inspiration comes from the FT(feature tree) in BIRCH(Balanced Iterative Reducing and Clustering using Hierarchies),abnormity detection base on clustering analysis, related data modal research in Data Base. After the analysis of property information,consuming bills and calling behavior records of telecom subscribers, initial fraud features are proposed and will be modified by further analysis. These features buildup a fraud feature tree, every node of the tree is a justicer, who determines which way you come in next step or how your fraud value is going to be modified. After going through all these feature nodes on the tree, each subscriber gets a fraud value which reflects how mach possibility the subscriber has in telecom fraud behavior.The creative points in this paper also include the data analysis methods. We use efficient data process tools, such as MATLAB and office data analysis tools. They turn the illogic discrete and meaningless data to vivid graphs that reveal the latent deception in telecom subscribers'behavior. Besides, Eclips and oracle tools are used to deal with the system procedure.