Telecommunications Save Card Research And Analysis Based On Hadoop

As the telecom industry agents growing, phenomenon of illegal operations agents appearing in the market has become increasingly. Where in the main agent of irregularities, the establishment of "cat pool" equipments support card to take telecom operators agent remuneration. To combat such acts of agents, telecom operators analyze the development of new users per month whether support card users, however, this usually takes a lot of manpower and resources, and highest accuracy rate is only 85% by conventional databases. In order to distinguish support card users more efficient, this paper uses data mining techniques and Hadoop platform.In this paper, with the real data coming from a Jiangsu telecom operator during author’s internship, we research and analysis how to distinguish suport card users. Data mining commonly used BP neural network algorithm and K_Means algorithm are applied to support card analysis. According to K_Means algorithm simulation results, we find a relatively strong impact factor and modifying the original data by property enhancement, this method can improve the clustering results of K_Means algorithm, also the accuracy of the card support analysis. In order to improve K_Means analysis algorithm base on Canopy clustering algorithm, we choose two maximum distance Canopies as the initial cluster centers, this can improve clustering efficiency. Finally, we make Hadoop-based platform integrated into the Web platform which includes type conversion algorithm calls, task monitoring, results of inquiries and other functions, this improves the efficiency of the support card analysis. The results show that the proposed Hadoop-based analysis to support card in the correct rate and spending time with good improvement, compared with traditional telecom operators database analysis, the correct rate increases 6.32%; compared with the traditional K_Means cluster analysis algorithm to support card, the time spent reduces 64.22% by Hadoop with four nodes.This study can not only provide theoretical support for telecom operators to analysis support card, but also provide a new idea for other industries in fraud analysis.