The Telecommunications Data Mining Technology Research

China's telecom operators begin to self-reform and develop in order to catch up rapidly changing external environment, at the same time,competition between telecom operators has evolved more and more intensely, but it also brings more opportunities for development. Telecom operators have accumulated vast amounts of customer consumption data, in which hide knowledge can guide more accurate business decisions. So finding hidden valuable information in large amounts of data has great significance for today's telecom operators to serve customers better in competition. Data mining is to help people free from traditional analysis of numerous data, and provide another more intelligent, efficient technical method to get knowledge from data.This paper focuses on two parts, including fuzzy association and periodic data clustering. First part analysis hard problems of range divide in quantitative properties, then introduce fuzzy association rules. This paper summaries and comparatively analysises algorithms include:fuzzy association rules algorithm based on AprioriTid(FAMA), based on FP-Tree(FFP-Tree) and based on (FMA-LL), and compares their advantages and disadvantages between these algorithms with examples, verifies that FMA-LL has comprehensive ability in dealing with fuzzy attributes and efficiency in algorithm computing.Then based on FMA_LL algorithm,solve the problem of determining the membership function from both linear and nonlinear aspects.CURE algorithm constitutes fuzzy sets according to clustering center, GA algorithm encodes parameters of membership function to chromosome and do genetic manipulation to get fuzzy sets, then determine the linear function.FCM and GA_FCM algorithm do clustering on each attribute to get membership matrix,then approximate normality ambiguity function according to clustering center and membership matrix.Do experiments using adult dataset to analyze, proof advantages and disadvantages of algorithms.Experiments proof CURE,GA and GA_FCM have its merits and faults on membership and algorithm efficiency.Second part analysis the existing clustering algorithm for the time-series data.The existing clustering algorithm for the time-series data is improved from the similarity measure aspects, but the actual time-series data tend to have a certain periodicity and continuity. The existing algorithms often ignore the impact of periodicity and continuity on time-series data clustering algorithms. This issue is researched in this paper and extension method is used to solve this problem. Preliminary results show the feasibility and effectiveness.