Modified Energy Statistic for Unsupervised Anomaly Detection

For prognostics in industrial applications, the degree of anomaly of a test point from a baseline cluster is estimated using a statistical distance metric. Among different statistical distance metrics, energy distance is an interesting concept based on Newton’s Law of Gravitation, promising simpler...

Full description

Saved in:
Bibliographic Details
Published inInternational journal of prognostics and health management Vol. 12; no. 1
Main Author Mukherjee, Rupam
Format Journal Article
LanguageEnglish
Published The Prognostics and Health Management Society 01.01.2021
Subjects
anomaly detection
energy distance
mahalanobis distance
statistical distribution
Online AccessGet full text

Cover

More Information
Summary:For prognostics in industrial applications, the degree of anomaly of a test point from a baseline cluster is estimated using a statistical distance metric. Among different statistical distance metrics, energy distance is an interesting concept based on Newton’s Law of Gravitation, promising simpler computation than classical distance metrics. In this paper, we review the state of the art formulations of energy distance and point out several reasons why they are not directly applicable to the anomaly-detection problem. Thereby, we propose a new energy-based metric called the P-statistic which addresses these issues, is applicable to anomaly detection and retains the computational simplicity of the energy distance. We also demonstrate its effectiveness on a real-life data-set.
ISSN:2153-2648
2153-2648
DOI:10.36001/ijphm.2021.v12i1.1323