Improve the Detection of Clustered Outliers via Outlier Score Propagation
Yongmou Li, Yijie Wang, Hongtao Guan · 2019
Most of the existing outlier detection methods in the literature often adopt the assumption that the outliers would locate in the low-density area of the input data space. However, such low-density assumption may not hold in the presence of clustered outliers, which usually demonstrate relatively high density. For this reason, the clustered and cluster-based outlier detection methods are proposed; however, these methods either make strong data distribution assumptions or rely on the high-quality clustering results which are often not available. To address this issue, we propose a graph-based outlier detection approach which is capable of significantly enhancing the performance of existing outlier detection methods, in the presence of clustered outliers. Experiments are conducted on both synthesized and real-world datasets to demonstrate the superiority of our proposed method. In particular, our proposed method outperforms the state-of-the-art clustered outlier detection methods by a large margin.