Sparse random projection isolation forest for outlier detection

Isolation Forest has a low computational complexity, hence has been widely applied to detect outliers in large-scale data. However, it suffers from the artifacts caused by the hyperplanes chosen, thereby failing to detect outliers in some specific regions. To tackle this problem, we propose the random-projectionbased Isolation Forest, which works in two steps. First, we transform the data using the random projection technique. Then, we employ the Isolation Forest to identify outliers using the transformed data. Experimental results show that the proposed methods outperform 12 state-of-the-art outlier detectors.(c) 2022 Published by Elsevier B.V.

Automated summary

Isolation Forest is widely used for outlier detection in large-scale data due to its low computational complexity. However, it may fail to detect outliers in specific regions due to artifacts caused by the chosen hyperplanes. To address this issue, a random-projection based Isolation Forest is proposed, which transforms the data and improves outlier detection performance.