On the Evaluation of Unsupervised Outlier Detection:
Measures, Datasets, and an Empirical Study
This webpage presents the supplementary material for the paper
On the Evaluation of Unsupervised Outlier Detection: Measures, Datasets, and an Empirical Study
by G. O. Campos, A. Zimek, J. Sander, R. J. G. B. Campello, B. Micenková, E. Schubert, I. Assent and M. E. Houle
Data Mining and Knowledge Discovery 30(4): 891-927, 2016, DOI: 10.1007/s10618-015-0444-8
We provide all datasets together with their descriptions here as well as all results visualized in graphs.
Since we plan on building a larger, and updated repository, the original results can be found in the DAMI results folder.
Changes to the published version:
The results of DAMI include an additional evaluation metric, denoted as
This measure is obtained by taking the maximum F-measure (harmonic mean of precision and recall) over all possible outlier score thresholds.
We include ODIN with k=1. Because of the change in the definition of the kNN, this parameter is now valid (previously, every point would have been its own 1-nearest neighbor).
No additional results have been added yet.
To archive these results long-term, we aim at hosting multiple copies at different universities. Currently, the following mirrors are available:
- Database Group, Ludwig-Maximilians-Universität München, Europe, primary site.
- Data Pattern Analysis Lab. University of São Paulo at São Carlos, South America