Improving angle based mappings

Abstract

Visualization of high-dimensional data is an important issue in data mining as it enhances the chance to selectively choose appropriate techniques for analyzing data. In this paper, two extensions to recent angle based multi-dimensional scaling techniques are presented. The first approach concerns the preprocessing of the data with the objective to lower the error of the subsequent mapping. The second aims at improving differentiability of angle based mappings by augmenting the target space by one additional dimension. Experimental results demonstrate the gain of efficiency in terms of layout quality and computational complexity. © 2008 Springer-Verlag Berlin Heidelberg.