SVDD regularized with area under the ROC

Authors:
Chen Bin;Li Bin;Pan Zhisong
Affiliations:
Yangzhou University, Yangzhou, China;Yangzhou University, Yangzhou, China;PLA University of Science Technology, Nanjing, China
Venue:
Proceedings of the 2nd International Conference on Interaction Sciences: Information Technology, Culture and Human
Year:
2009

Citing 4
Cited 0

The nature of statistical learning theory

The nature of statistical learning theory
Support vector domain description

Pattern Recognition Letters - Special issue on pattern recognition in practice VI
Novelty detection: a review—part 1: statistical approaches

Signal Processing
Support Vector Data Description

Machine Learning

Quantified Score

Hi-index	0.00

Visualization

Abstract

Support Vector Data Description (SVDD) has been successfully applied to detect outliers as points outside the minimal ball enclosing given target training objects. In its design, some rarely-happened (labeled) outliers can be available, SVDD using Negative examples (NSVDD) can develop a more exact description than SVDD by utilizing the outliers to minimize the empirical risk and thus obtain better classification. However, the low classification error does not always mean good ROC performance when facing the problems with the unknown misclassification costs and/or the class distributions. Inspired by the characteristic of Area Under ROC (AUC) against the above problems, in this paper, we develop a SVDD regularized by the AUC (RSVDD) to optimize the empirical risk added a AUC regularizer in presence of a few available outliers. Experimental results verify the effectiveness of RSVDD in comparison with the traditional SVDD, NSVDD, ROC optimizer Support Vector Machine (ROCSVM) and SVM.