The Research Of Distributed Data Mining Based On M-Agent System

With the development of the computer network and distributed database mining techniques, searching knowledge from the distributed system has great application prospects. The issues of extacting knowledge from distributed database by traditional centralized data mining methods are often not well resolved with the huge amount of data. Since:①the constraints of the Network bandwidth;②using traditional data mining methods doesn't work effectively on centralized data;③the data usually not allowed moving in security aspects. Therefore, how to extract knownledge from the distributed system efficiently has become researching hotspots currently.Since 1990s, Multi-Agent System(MAS) is gradually becoming a research hotspot and has aprofound impact on the development of science and technology.The Multi-Agent technology is suitable for the complex distributed problems.In Data Mining System based on Multi-Agent, Agents on different database will be working according to the user's requirements, and transferring the result to the users. This method can not only avoid the mass movement of data, but also can solve the security problem. So there are advantages of using the Multi-Agent technology in the distributed database mining system.This paper represents the principle of Data Mining based on Multi-Agent technology, then designed a distributed data mining system model based on MAS, discussed the processes of the system, and established the structure and function of various Agents.In order to improve the performance of distributed association rule mining, this paper designed a weighted frequent itemsets incremental updating algorithm(WHIUA) based on hash table, and the improved FP-growth algorithm was applied to the distributed system. Analyses and experiments show that these efforts can not only deal with the problem that database and minimum support are both changed, but also can effectively increases the excavating and updating efficiency of maximum frequent itemsets, and then improve the performance of distributed data mining.