How can we evaluate the performance of an unsupervised anomaly detector
For offline approaches ROC Curve and for online approaches AMOC curve
Hi Muhammed,
Here are some references on the subject :
- the authors present in the first paper a new geometric framework for unsupervised anomaly detector which are algorithms that are designed to process unlabled data.
- In the second paper the authors present the problem of detecting anomalies in time series data using Peer Group Analysis (PGA), which is an unsupervised technique.
- Finally, in the third paper the authors present the semi-supervised anomaly detection needs to ground on the unsupervised learning paradigm and devise a novel algorithm that meets this requirement.
1) A Geometric Framework for Unsupervised Anomaly Detection:Anomaly Detection Using Unsupervised Profiling
www.cs.cmu.edu/~aarnold/cald/uad-dmsa02.pdf
2) Anomaly Detection Using Unsupervised Profiling Method in Time Series Data
ceur-ws.org/Vol-215/paper06.pdf
3) Toward Supervised Anomaly Detection
https://www.jair.org/media/.../live-3623-6845-jair.pdf
Good Luck
Hi Muhammed,
Another reference that may interest you
Histogram-based OutlierScore(HbOS) : A fast Unsupervised Anomaly Detection Algorithm
WWW.dfki.de/k12012/.../ki2012pd13.pdf
Good Luck
Hi Muhammed,
For real time application, time response and accuracy (%, precision) are most important.
Good Luck
Hi Muhammed
1- For real time application, time response and accuracy .
2-For offline approaches( ROC )Curve and for online approaches( AMOC )curve.
Good Luck
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory
- Control Systems
- Control Systems Engineering