Instance space analysis for unsupervised outlier detection

Sevvandi Kandanaarachchi, Mario A. Muñoz, Kate Smith-Miles

Research output: Chapter in Book/Report/Conference proceedingConference PaperOtherpeer-review

Abstract

The success of various methods for unsupervised outlier detection t depends on how well their definition of an outlier matches the dataset properties. Given that each definition, and hence each t method, has strengths and weaknesses, measuring those properties could help us to predict the most suitable method for the dataset at hand. In this paper, we construct and validate a set of meta-features that measure such properties. We then conduct the first instance space analysis for unsupervised outlier detection methods based on the meta-features. The analysis provides insights into the methods’ strengths and weaknesses, and facilitates the recommendation of an appropriate method with good accuracy. (

Original languageEnglish
Title of host publicationProceedings of the 1st Workshop on Evaluation and Experimental Design in Data Mining and Machine Learning co-located with SIAM International Conference on Data Mining (SDM 2019)
EditorsEirini Ntoutsi, Erich Schubert, Arthur Zimek, Albrecht Zimmermann
Place of PublicationGermany
PublisherRheinisch-Westfaelische Technische Hochschule Aachen
Number of pages9
Publication statusPublished - 2019
EventEvaluation and Experimental Design in Data Mining and Machine Learning 2019 - Hyatt Regency Calgary, Calgary, Canada
Duration: 4 May 2019 → …
Conference number: 1st
https://imada.sdu.dk/Research/EDML/ (Workshop website)
http://ceur-ws.org/Vol-2436/ (Proceedings)

Publication series

NameCEUR Workshop Proceedings
Volume2436
ISSN (Print)1613-0073

Workshop

WorkshopEvaluation and Experimental Design in Data Mining and Machine Learning 2019
Abbreviated titleEDML 2019
CountryCanada
CityCalgary
Period4/05/19 → …
OtherWorkshop at the SIAM International Conference on Data Mining (SDM19)
Internet address

Cite this