Distributed Data Mining Based On Grid Services

The main purpose of distributed data mining is to reduce the cost of network communication and use the computing resource and storage resource which are geographically distributed. This thesis adopts grid technique to incorporate this geographically distributed computing resource and execute the distributed data mining, and then the purpose is implemented.First, this thesis introduces some main native and foreign grid projects and grid commodity, then explain some basic concept of grid. Next, compare grid with some traditional distributed techniques. It expatiate prevail grid architecture, and show the basic concept and correlative performance parameter definition of association rules. Analyze distributed association rules mining algorithm ODAM base on researching current prevail parallel and distributed data mining algorithm. The thesis proposes three improvements of ODAM algorithm which show as follow. First one: it judge whether or not the number of n-1 global frequent itemsets is less than n before generating n candidate itemsets, if yes then the mining is end. Second one: judge whether or not n is bigger than the max number of item count in a transaction, if yes then the mining is over. Third one: every site calculates part candidate itemsets, and then unites this local candidate from every site to make a global itemsets.The Thesis use radar data set and connect-4 data set do experiment, and the results prove that the three improvements are effective.Finally, the thesis discuss the implements of the improved ODAM algorithm base on grid service in detail with GT3, and does experiment by adopting radar data and connect-4 data set, experiment proved the idea that grid can incorporate computing ability.