The Research Of Data Mining Based On Enterprise Data Warehouse Of Telecom

The business of telecom is the fastest-rising in the global economy in recent years and it is also the most hotly competitive at the same time. How telecom gets consumers to favor in numerous enterprises to improve the analytical capacity and the market competitiveness of enterprises and maintains the leading position of market is the most severe test at present.The aggravation of the competition makes the data warehouse analysis as the data platform supported to make policy prevails day by day, and data mining at the platform is increasingly used.Firstly, in this thesis we explain the concept of data warehouse and data mining and introduce the construction of the enterprise data warehouse. Then we propose the enterprise data warehouse of the telecom system. Secondly, we propose the module design and realization of data mining based on enterprise data warehouse which combine the data mining module and enterprise data warehouse to meet the demand of data mining for telecom.We propose Hash Partition Algorithm and Genetic Algorithm based on K-means according to association analysis of call mode of customers and clustering of VIP after carrying on deep research to enterprise datum warehouse.Hash Partition Algorithm takes performance of data mining into consideration on the basis of mass data of Telecom. It has realized reducing the number of times of scanning database greatly by partition design of data. At the same time, having realized the integration of hash algorithm and partition algorithm, it provides the steps of realization and the experimental result based on enterprise data warehouse.Genetic algorithm based on K-means realizes the hybrid of genetic algorithm and K-means algorithm which is suitable for mass data mining. Thealgorithm can obviously reduce the number of scanning database, improving the algorithm performance and can fully reflect the characteristic of the VIP customers to realize the cluster to VIP customers' characteristics.