Multidimensional Scaling Visualization Using Parametric Entropy

Abstract

This paper studies complex systems using a generalized multidimensional scaling (MDS) technique. Complex systems are characterized by time-series responses, interpreted as a manifestation of their dynamics. Two types of time-series are analyzed, namely 18 stock markets and the gross domestic product per capita of 18 countries. For constructing the MDS charts, indices based on parametric entropies are adopted. Multiparameter entropies allow the variation of the parameters leading to alternative sets of charts. The final MDS maps are then assembled by means of Procrustes’ method that maximizes the fit between the individual charts. Therefore, the proposed method can be interpreted as a generalization to higher dimensions of the standard technique that represents (and discretizes) items by means of single “points” (i.e. zero-dimensional “objects”). The MDS plots, involving one-, two- and three-dimensional “objects”, reveal a good performance in capturing the correlations between data.