关键词: 离群点, 聚类, 阴影集

Abstract: This paper proposes a new definition for outliers from the macroscopic characteristics of data sets, and designs a new outlier factor of observation (COF) by considering both deviation of outlier to clusters and uncertainty of outliers itself. The paper gives a new optimization goal on Shadowed Sets, which pays more attention to the accuracy of core in the shadowed process of fuzzy sets. Further, the paper develops an outlier detection algorithm based on Shadowed Sets clustering to incorporate the advantages of both COF and Shadowed Sets in a hybridized framework. The experimental results on synthetic and Iris data sets demonstrate better effect of the proposed approach.

Key words: outlier, clustering, Shadowed Sets