aalto1 untyped-item.component.html

Semi-automated labeling of time series for anomaly detection

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Department

Major/Subject

Mcode

SCI3115

Language

en

Pages

42

Series

Abstract

Anomaly detection in Time Series is a widespread topic because there is a huge amount of Time Series data, and anomalies in them indicate possible important events. However, labeled data sets are scarce resource, thus studying and implementing automated or semi-automated labeling algorithms for anomalies in Time Series is a promising research area. This work presents a semi-automated labeling technique for anomaly detection in Time Series. I propose a tool that extracts most prominent parts of a series as a candidates to label as normal or anomalous. Additionally, I integrated this tool into existing Time Series analysis system. The goal is to present to human annotator the most representative parts of the series, but in a reasonable amount, so that human expert can decide which area is of real interest. We can further search for areas similar to those confirmed by human expert and propagate corresponding labels, so that time and effort are minimized.

Description

Supervisor

Riveill , Michel

Thesis advisor

Muller, Julien

Other note

Citation

Endorsement

Review

Supplemented By

Referenced By