关键词: 数据挖掘, 离群数据, 网格, 数据分区, 单元, 微单元, 维单元树

Abstract: To solve the problems of inefficiency and bad-adaptability for the existing outlier mining algorithms based on grid, this paper proposed an outlier mining algorithm based on data partitioning and grid. Firstly, the technology of data partitioning was applied. Secondly, the non-outliers were filtered out by cell and the intermediate results were temporarily stored. Thirdly, the structure of the improved Cell Dimension Tree (CD-Tree) was created to maintain the spatial information of the reserved data. Afterwards, the non-outliers were filtered out by micro-cell and were operated efficiently through two optimization strategies. Finally, followed by mining by data point, the outlier set was obtained. The theoretical analysis and experimental results show that the method is feasible and effective, and has better scalability for dealing with massive and high dimensional data.

Key words: data mining, outlier data, grid, data partitioning, cell, micro-cell, Cell Dimension Tree (CD-Tree)