A New Data Description Method Based on the Density-Induced Support Vector Information
Kyung‐Yul Lee, Doheon Lee, Kwang-Hyung Lee · 2005
We propose a new support vector data description (SVDD) methodology wherein we incorporate the density distribution of a training data set by introducing a relative density degree for each data point. By using the density-induced distance measurement, we generalize a conventional SVDD in order to reflect the density of given data. Experiments with various real data sets show that the proposed method describes more accurately given data sets than other current methods including the conventional SVDD. We refer to the proposed method as a Density-Induced Support Vector Data Description.