Abstract: An algorithm on cell-based outlier detection is put forward. The data spatial cell partitioning and data object allocation methods in cell outlier detection algorithms are discussed. Algorithm is improved to overcome the limitation of requiring threshold. Using the data of taxpaying behavior. Compared with other algorithm, result shows that this approach can not only detect the outliers efficiently, but also decide the position of outliers.

Key words: data mining, outliers detection, cell, taxpaying behavior

摘要： 研究基于单元的孤立点检测算法，给出数据空间的单元格划分及数据对象分配算法。针对该算法中阈值M设置的不足，对算法进行改进并应用于纳税行为的分析。与其他孤立点检测算法对比的结果表明，该算法不仅能有效挖掘纳税行为中的孤立点，还能确定孤立点的位置，有利于对纳税行为的分析。

关键词: 数据挖掘, 孤立点检测, 单元, 纳税行为

CLC Number: 

• TP301.6