Title: Robust novelty detection in the framework of a contamination neighbourhood

Authors: Lev V. Utkin; Yulia A. Zhuk

Addresses: Department of Control, Automation and System Analysis, St. Petersburg State Forest Technical University, Institutsky per. 5, 194021 St. Petersburg, Russia ' Department of Computer Science, St. Petersburg State Forest Technical University, Institutsky per. 5, 194021 St. Petersburg, Russia

Abstract: A novelty detection robust model is studied in the paper. It is based on contaminated (robust) models which produce a set of probability distributions of data points instead of the empirical distribution. The minimax and minimin strategies are used to construct optimal separating functions. An algorithm for computing the optimal parameters of the novelty detection model is reduced to a finite number of standard SVM tasks with weighted data points. Experimental results with synthetic and some real data illustrate the proposed novelty detection robust model.

Keywords: machine learning; novelty detection; classification; minimax strategy; support vector machine; SVM; quadratic programming; probability distribution; weighted data points; modelling; unknown data.

DOI: 10.1504/IJIIDS.2013.053830

International Journal of Intelligent Information and Database Systems, 2013 Vol.7 No.3, pp.205 - 224

Received: 05 Apr 2012
Accepted: 02 Nov 2012
Published online: 31 Mar 2014 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article