Angle-Based Outlier Detection Algorithm with More Stable Relationships

Abstract

Outlier detection is very useful in many applications, such as fraud detection and network intrusion. The angle-based outlier detection (ABOD) method, proposed by Kriegel, plays an important role in identifying outliers in high-dimensional spaces. However, ABOD only considers the relationships between each point and its neighbors and does not consider the relationships among these neighbors, causing the method to identify incorrect outliers. In this paper, we provide a small but consistent improvement by replacing the relationships between each point and its neighbors with more stable relationships among neighbors. Compared with other related methods, which work best in either high or low-dimensional spaces, our method gives significant gains in both high and low-dimensional spaces. Experimental results on both synthetic and real-world datasets demonstrate the effectiveness of our method.