Skip to Main content Skip to Navigation
Conference papers

Human-Interpretable Rules for Anomaly Detection in Time-series

Abstract : Anomaly detection in time series is a widely studied issue in manyareas. Anomalies can be detected using rule-based approachesand human-interpretable rules for anomaly detection refer torules presented in a format that is intelligible to analysts. Learningthese rules is a challenge but only a few works address the issue ofdetecting different types of anomalies in time-series. This paperpresents an extended decision tree based on patterns to generatea minimized set of human comprehensible rules for anomalydetection in univariate times-series. This method uses Bayesianoptimization to avoid manual tuning of hyper-parameters. Wedefine a quality measure to evaluate both the accuracy and theintelligibility of the produced rules. Experiments show that ourapproach generates rules that outperforms the state of- the-artanomaly detection techniques.
Complete list of metadata
Contributor : Ines Ben Kraiem <>
Submitted on : Thursday, April 22, 2021 - 2:23:16 PM
Last modification on : Wednesday, June 9, 2021 - 10:00:32 AM


Publisher files allowed on an open archive


Distributed under a Creative Commons Attribution 4.0 International License



Inès Ben Kraiem, Faiza Ghozzi, André Péninou, Geoffrey Roman-Jimenez, Olivier Teste. Human-Interpretable Rules for Anomaly Detection in Time-series. 24th International Conference on Extending Database Technology (EDBT 2021), University of Cyprus, Mar 2021, Nicosia (virtual), Cyprus. pp.457-462, ⟨10.5441/002/edbt.2021.51⟩. ⟨hal-03205628⟩



Record views


Files downloads