Visual evaluation of outlier detection models

Elke Achtert*, Hans Peter Kriegel, Lisa Reichert, Erich Schubert, Remigius Wojdanowski, Arthur Zimek

*Kontaktforfatter for dette arbejde

Publikation: Kapitel i bog/rapport/konference-proceedingKonferencebidrag i proceedingsForskningpeer review

Abstrakt

Many outlier detection methods do not merely provide the decision for a single data object being or not being an outlier. Instead, many approaches give an "outlier score" or "outlier factor" indicating "how much" the respective data object is an outlier. Such outlier scores differ widely in their range, contrast, and expressiveness between different outlier models. Even for one and the same outlier model, the same score can indicate a different degree of "outlierness" in different data sets or regions of different characteristics in one data set. Here, we demonstrate a visualization tool based on a unification of outlier scores that allows to compare and evaluate outlier scores visually even for high dimensional data.

OriginalsprogEngelsk
TitelDatabase Systems for Advanced Applications - 15th International Conference, DASFAA 2010, Proceedings
RedaktørerH. Kitagawa, Y. Ishikawa, Q. Li, C. Watanabe
ForlagSpringer
Publikationsdato28. dec. 2010
UdgavePART 2
Sider396-399
ISBN (Trykt)978-3-642-12097-8
ISBN (Elektronisk)978-3-642-12098-5
DOI
StatusUdgivet - 28. dec. 2010
Udgivet eksterntJa
Begivenhed15th International Conference on Database Systems for Advanced Applications - Tsukuba, Japan
Varighed: 1. apr. 20104. apr. 2010

Konference

Konference15th International Conference on Database Systems for Advanced Applications
LandJapan
ByTsukuba
Periode01/04/201004/04/2010
SponsorKIISE Database Society of Korea, China Computer Federation Database Technical Committee, ARC Research Network in Enterprise Information Infrastructure, Asian Institute of Technology, Information Processing Society of Japan (IPSJ)
NavnLecture Notes in Computer Science
Vol/bind5982
ISSN0302-9743

Fingeraftryk Dyk ned i forskningsemnerne om 'Visual evaluation of outlier detection models'. Sammen danner de et unikt fingeraftryk.

Citationsformater