Feature scaling in support vector data description

Abstract

When in a classification problem only samples of one class are easily accessible, this problem is called a one-class classification problem. Many standard classifiers, like backpropagation neural networks, fail on this data. Some other classifiers, like k-means clustering or nearest neighbor classifier can be applied after some minor changes. In this paper we focus on the support vector data description classifier, which is especially constructed for one-class classification. But this method appears to be sensitive to scaling of the individual features of the dataset. We show that it is possible to improve its performance by adequate scaling of the feature space. Some results will be shown on artificial dataset and handwritten digits dataset.