Abstract:
Anomaly detection plays an increasingly important role in various fields for critical tasks such as intrusion detection in cybersecurity and financial risk detection. The isolation forest based anomaly detection methods with iForest as the representative have shown excellent detection performance with very fast execution speed. However, iForest and its variants still suffer from several drawbacks such as lack of theoretical foundations for the effectiveness of the isolation forest mechanism, limited applicability to advanced data types and complicated anomalies, and difficulty in working with the deep learning scheme. In this talk, we would like to introduce a series of works to address the limitations in existing isolation forest based anomaly detection methods. Firstly, we theoretically link the locality-sensitive hash techniques with isolation forest and propose a generic isolation forest based anomaly detection framework LSHiForest. Then, we further theoretically analyse how the branching factor affective the effectiveness of isolation and establish the notion of optimal isolation forest for anomaly detection. Finally, we show how to deepen and optimise the isolation forest mechanism with evolutionary algorithms to achieve deep isolation forest for anomaly detection.
Page(s):
1-1
DOI:
DOI not available
Published:
Journal: Second International Conference on Computing Technologies, Tools and Applications (ICTAPP-24), June 4-6,2024 (Abstract Book), Volume: 0, Issue: 0, Year: 2024