Abstract: According to the measurement results in high dimensional data is not obvious problems of the traditional distance, proposes a shared nearest neighbor subspace clustering algorithm （SNN_SC）, according to the dimension of the data set into multiple nearest neighbor transaction database mining in transaction database maximum co-occurrence object set, namely dimension clustering. On the one dimensional clustering set, the closed frequent itemsets are further exploited. The dimension of the closed frequent itemsets is a subspace. The experimental results show that SNN_SC can more accurately locate the subspace, and generate a complete clustering in subspace.

Key words: high dimensional, shared nearest neighbor, subspaceclustering, closed frequent items

摘要： 针对传统距离度量在高维数据上效果不明显问题，提出一种共享最近邻子空间聚类算法（SNN_SC），按照维把数据集转变为多个最近邻事务数据库，挖掘事务数据库中最大共现对象集，即一维上聚类。在一维聚类集上进一步挖掘闭频繁项集，包含闭频繁项集的维是子空间，闭频繁项集是子空间上聚类。实验对比结果表明，SNN_SC能够更准确定位子空间，并在子空间上产生完整聚类。

关键词: 高维, 共享最近邻, 子空间聚类, 闭频繁项集