|
Performance assessment of outlier detection algorithms in time series
|
S. Zaminpardaz *   , M. A. Sharifi , A. R. Amiri-Simkooei |
|
|
|
Abstract: (6937 Views) |
| Our purpose in this contribution is to compare different outlier detection methods as far as time series are concerned. In fact, three methods, namely wavelet analysis, Baarda data snooping and thresholding are investigated. In order to make reasonable comparisons among the performance of these three methods in detecting the outliers, we used 4-month synthesized time series based on real tidal data. When the functional model of observations is known, Baarda data snooping, in comparison with other two methods yields the best results since its outlier rate of success and outlier rate of failure are almost 100% and 0.64%, respectively, regardless of the type of outliers. Furthermore, if the functional model of observations is unknown, wavelet analysis perform better than thresholding. |
|
| Keywords: Outlier Detection, Wavelet Analysis, Baarda Data Snooping, Thresholding, Time Series |
|
|
Full-Text [PDF 542 kb]
(2263 Downloads)
|
Type of Study: Research |
Subject:
Geo&Hydro
Received: 2015/06/13
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
