Robust Anomaly Detection on Unreliable Data

Abstract : Abstract—Classification algorithms have been widely adoptedto detect anomalies for various systems, e.g., IoT and cloud, underthe common assumption that the data source is clean, i.e., featuresand labels are correctly set. However, data collected from the fieldcan be unreliable due to careless annotations or malicious datatransformation for incorrect anomaly detection. In this paper,we present a two-layer learning framework for robust anomalydetection (RAD) in the presence of unreliable anomaly labels.The first layer of quality model filters the suspicious data,where the second layer of classification model detects the anomalytypes. We specifically focus on two use cases, (i) detecting 10classes of IoT attacks and (ii) predicting 4 classes of task failuresof big data jobs. Our evaluation results show that RAD canrobustly improve the accuracy of anomaly detection, to reach upto 98% for IoT device attacks (i.e., +11%) and up to 83% forcloud task failures (i.e., +20%), under a significant percentage ofaltered anomaly labels.
Type de document :
Communication dans un congrès
49th IEEE/IFIP International Conference on Dependable Systems and Networks (DSN 2019), Jun 2019, Portland, Oregon, United States
Liste complète des métadonnées

https://hal.archives-ouvertes.fr/hal-02056558
Contributeur : Zilong Zhao <>
Soumis le : lundi 4 mars 2019 - 16:35:14
Dernière modification le : mercredi 13 mars 2019 - 13:36:47

Identifiants

  • HAL Id : hal-02056558, version 1

Citation

Zilong Zhao, Sophie Cerf, Robert Birke, Bogdan Robu, Sara Bouchenak, et al.. Robust Anomaly Detection on Unreliable Data. 49th IEEE/IFIP International Conference on Dependable Systems and Networks (DSN 2019), Jun 2019, Portland, Oregon, United States. 〈hal-02056558〉

Partager

Métriques

Consultations de la notice

71