The Research Of Distributed Privacy Preserving Data Mining Based On Intelligent Agent

Since the1990s, Multi-Agent System has gradually become a research hotspot, and had aprofound influence on the development of a lot of science and technology. Multi-Agent Systemis more suitable for complicated distributed computing and can effectively avoid the problem ofdata movement at the large-scale distributed site. Therefore, applying the Multi-Agenttechnology to Distributed Data Mining has a great advantage, also can speed up the progress ofDistributed Data Mining technology.Firstly, this paper summarizes the significance of the research subject, and then makesanalysis and discussion of Distributed Data Mining, Privacy Preserving technology andDistributed Privacy Preserving Data Mining. Finally, the paper introduces the relevanttechnology about Multi-Agent and deeply makes study on one Multi-Agent System frameworkbased on Java, namely JAFMAS. By expanding the JAFMAS, a prototype system applied toDistributed Privacy Preserving Data Mining is designed and implemented. The main work of thispaper includes as follows.(1) For the defects and deficiencies existing in Secure Multi-Party Computation (SMC)technology applied to the large-scale distributed data environment, an improved Apriorialgorithm based on distributed Association Rules Mining is given. It can reduce the times ofdatabase scanning and the amount of calculation of candidate frequent itemsets, and it can beverified.(2) In the Multi-Agent System, communication is the basis of cooperation, so one bestcommunication mechanism should be chosen to realize the mutual cooperation between multipleAgents, and to finish the system task by make coordination of multiple Agents. This paperproposes an Agent communication mechanism by combination of point to point communicationand new broadcasting, according to the relevant support provided by the communication protocol layer of JAFMAS framework, and gives an improved communication model which canbeneficially improve the mechanism of coordination and cooperation.(3) For the proposed Distributed Privacy Preserving Data Mining framework model, thispaper has constructed a prototype system applied to Distributed Privacy Preserving Data Miningby expanding and implementing the JAFMAS framework, and has given the design of eachfunction module and analysis of process.Finally, this paper makes related test on the prototype system, mainly to analyse theimproved Association Rules Mining algorithm and the improved Agent communicationmechanism. The experimental results show that the prototype system has a great degree ofimprovement and perfection, can better meet the needs of Distributed Privacy Preserving DataMining and has improved the entire performance.