With the development of informatisation, the information data is distributed to different departments and every department has the need to fully cooperate with each other in condition of its own information not being leaked;on the other hand,the concentrated calculation cannot satisfy the requirement of different application due to huge amount of the information data. The distributed data mining becomes one of the research hot-points in above background. In this paper, by dividing the system into core-nodes and periphery-nodes,we conduct the hierarchical management and reduce system' s burden brought by the communication of information. The definition of micro-clustering will be presented in the paper and the algorithm is described in the periphery-nodes. Experiment illuminate that our distributed algorithm has similar accuracy rate as that of the concentrated K-means algorithm in condition of assuring no leakage of every department' s data. This has demonstrated the feasibility and validity of the algorithm.%由于信息化的不断拓展,一方面数据广泛分布于不同的部门,各部门需求在不泄漏自身数据的情况下进行充分合作;另一方面由于数据量的巨大,集中式的运算已经很难满足各种应用的需求.在这一个背景下,分布式数据挖掘已经成一个研究的热点,通过将系统划分为中心节点和边缘节点,分层管理并减少数据交流给系统带来的负担;文章还将给出微簇的定义,并在边缘节点中进行算法的描述.实验说明在保证各部门的数据不泄漏的情况下,分布式算法具有跟集中式的K-means算法相近的准确率,说明算法具有可行性和有效性.
展开▼
