Outlier detection and quasi-periodicity optimization algorithm: Frequency domain based outlier detection (FOD)

Erkuş E. C., Purutcuoglu V.

EUROPEAN JOURNAL OF OPERATIONAL RESEARCH, vol.291, no.2, pp.560-574, 2021 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 291 Issue: 2
  • Publication Date: 2021
  • Doi Number: 10.1016/j.ejor.2020.01.014
  • Journal Name: EUROPEAN JOURNAL OF OPERATIONAL RESEARCH
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, International Bibliography of Social Sciences, ABI/INFORM, Applied Science & Technology Source, Business Source Elite, Business Source Premier, Computer & Applied Sciences, EconLit, INSPEC, Public Affairs Index, zbMATH, Civil Engineering Abstracts
  • Page Numbers: pp.560-574
  • Keywords: Fourier transform, Periodicity, Optimization, Outlier detection, PRINCIPAL COMPONENT ANALYSIS, SEASONAL SHIFT, TEMPERATURE, IDENTIFICATION
  • Middle East Technical University Affiliated: Yes

Abstract

Outlier detection is one of the main challenges in the pre-processing stage of data analyses. In this study, we suggest a new non-parametric outlier detection technique which is based on the frequency-domain and Fourier Transform definitions and call it as the frequency-domain based outlier detection (FOD). From simulation results under various distributions and real data applications, we observe that our proposal approach is capable of detecting quasi-periodic outliers in time series data more successfully compared with other commonly used methods like z-score, box-plot and also faster than some specialized methods Grubbs method and autonomous anomaly detection (AAD) method. Therefore, we consider that our proposal approach can be an alternative approach to find quasi-periodic outliers in time series data. (c) 2020 Elsevier B.V. All rights reserved.