FINDING DISTANCE-BASED OUTLIERS IN SUBSPACES THROUGH BOTH POSITIVE AND NEGATIVE EXAMPLES

Fabio Fassetti, Fabrizio Angiulli

2010

Abstract

In this work an example-based outlier detection method exploiting both positive (that is, outlier) and negative (that is, inlier) examples in order to guide the search for anomalies in an unlabelled data set, is introduced. The key idea of the method is to find the subspace where positive examples mostly exhibit their outlierness while at the same time negative examples mostly exhibit their inlierness. The degree to which an example is an outlier is measured by means of well-known unsupervised outlier scores evaluated on the collection of unlabelled data. A subspace discovery algorithm is designed, which searches for the most discriminating subspace. Experimental results show that the method is able to detect a near optimal solution, and that the method is promising from the point of view of the knowledge mined.

Paper Citation

in Harvard Style

Fassetti F. and Angiulli F. (2010). FINDING DISTANCE-BASED OUTLIERS IN SUBSPACES THROUGH BOTH POSITIVE AND NEGATIVE EXAMPLES . In Proceedings of the 2nd International Conference on Agents and Artificial Intelligence - Volume 1: ICAART, ISBN 978-989-674-021-4, pages 5-10. DOI: 10.5220/0002699600050010

in Bibtex Style

@conference{icaart10,
author={Fabio Fassetti and Fabrizio Angiulli},
title={FINDING DISTANCE-BASED OUTLIERS IN SUBSPACES THROUGH BOTH POSITIVE AND NEGATIVE EXAMPLES},
booktitle={Proceedings of the 2nd International Conference on Agents and Artificial Intelligence - Volume 1: ICAART,},
year={2010},
pages={5-10},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0002699600050010},
isbn={978-989-674-021-4},
}

in EndNote Style

TY - CONF
JO - Proceedings of the 2nd International Conference on Agents and Artificial Intelligence - Volume 1: ICAART,
TI - FINDING DISTANCE-BASED OUTLIERS IN SUBSPACES THROUGH BOTH POSITIVE AND NEGATIVE EXAMPLES
SN - 978-989-674-021-4
AU - Fassetti F.
AU - Angiulli F.
PY - 2010
SP - 5
EP - 10
DO - 10.5220/0002699600050010