Robust solutions to fuzzy one-class support vector machine

One-class SVM is used for classification which distinguishes one class of data from the rest in the feature space. For the training samples coming from different sources with different quality, in this letter, a reformulation of one-class SVM is proposed by simultaneously incorporating robustness and fuzziness to improve the classification performance. Based on the proposed model, we derive the relationship between the lower bound of fuzziness mu(min) and the upper bound of perturbation eta in the input data. Specifically, for a given eta, only when the assigned fuzziness to the input data is larger than mu(min) could the input data be in full use and differentiated effectively. The experiments verify the mathematical analysis and illustrate that the proposed model can achieve better classification performance. (C) 2016 Elsevier B.V. All rights reserved.