Kernel-based approach to anomaly detection

Project Title:

Kernel-based approach to anomaly detection

Supervisor:

Kai Ming Ting

Contact person and email address:

Kai Ming Ting: Kaiming.ting@federation.edu.au

Project Description:

There are existing kernel-based approaches in detecting anomalies in a dataset. However, they are inferior to density-based and isolation-based anomaly detectors in terms of detection accuracy and run time. This project aims to create a new kernel-based anomaly detector which outperforms not only existing kernel-based anomaly detectors but at least as good as existing state-of-the-art anomaly detectors, in terms of detection accuracy and run time. The appeal of kernel-based approach over density-based and isolation-based approaches is that it is more robust in high dimensional datasets. In addition, it will have the ability to detect point anomalies as well as group anomalies (where individual points in a group may be normal; but anomalous characteristic only reveals itself when the points can treated as a group.)