摘要： 针对传统离群点检测过程中属性多、维度大等问题，结合粗糙集理论，提出一种基于近邻关系的离群约简搜索算法。利用属性约简技术解决对象不相容的问题，并有效缩减离群搜索的属性空间。计算任意点与其他所有点间的距离和，通过计算基于近邻的加权离群因子来判定离群点，并在通用数据集上进行测试。实验结果表明，该离群检测算法的搜索精度较高。

关键词: 离群点检测, 离群因子, 核属性集, 决策表, 近邻关系

Abstract: Traditional outlier detection algorithms often take all attributes of the dataset into account thus result in heavy cost for handling high dimensional data. This paper proposes a searching algorithm for outlying reduction based on neighbor relation inspired by rough set theory and related techniques. This approach employs attribute reduction technology on the inconsistent decision table, which reduces the attribute fields. It computes the sum distance of the current point to all the other points and formulates a neighbor-based outlier factor to judge the abnormality of the data object. Experimental results on the public dataset show that this method is efficient and effective.

Key words: outlier detection, outlier factor, core attribute set, decision table, neighbor relation

中图分类号: 

• TP311