Publication detail

Outlier identification based on local extreme quantile estimation

HOLEŠOVSKÝ, J. KŮDELA, J.

Original Title

Outlier identification based on local extreme quantile estimation

Type

conference paper

Language

English

Original Abstract

An extensive time series observations serve for an input in wide range of technical, economical and environmental application areas. However, the verification of validity of such data is necessary condition for any further analysis. Correctness of the data can be proven with respect to various criteria, mainly the attention is focused on detecting possible outliers in the series. Among others, these comprise observations corrupted by failure of any measuring instrument or influence of other than the quantity of interest. In this contribution we present an advanced technique for time series outlier detection based on extreme value analysis. Extreme value theory is being successfully applied in many branches, and hence provides an adequate framework for detection of rare events such as outliers. The suitability of the method proposed is also discussed with respect to eventual automation of the whole procedure. The method was applied for validation of hourly air pollution data obtained in Brno, Czech Republic. The measurements were provided by automated instruments at locations with high traffic and industrial load. The proposed method might simplify the procedure of such extensive data verification.

Keywords

extreme value, outliers, return level, time series, heuristic optimization

Authors

HOLEŠOVSKÝ, J.; KŮDELA, J.

Released

8. 6. 2016

Publisher

Brno University of Technology

Location

Brno, Czech Republic

ISBN

978-80-214-5365-4

Book

Proceedings of 22nd International Conference on Soft Computing MENDEL 2016

Edition number

2016

ISBN

1803-3814

Periodical

Mendel Journal series

Year of study

2016

State

Czech Republic

Pages from

255

Pages to

260

Pages count

6

BibTex

@inproceedings{BUT126080,
  author="Jan {Holešovský} and Jakub {Kůdela}",
  title="Outlier identification based on local extreme quantile estimation",
  booktitle="Proceedings of 22nd International Conference on Soft Computing MENDEL 2016",
  year="2016",
  journal="Mendel Journal series",
  volume="2016",
  number="2016",
  pages="255--260",
  publisher="Brno University of Technology",
  address="Brno, Czech Republic",
  isbn="978-80-214-5365-4",
  issn="1803-3814"
}