A proposal for spatio-temporal analysis of traffic matrices using HJ-biplot

Abstract

Since 2002 many works have been published applying Principal Component Analysis (PCA) in the study of network traffic. These investigations have revealed some issues inherent to the temporal and spatial correlations present in the data, which are not considered in PCA. The proposed solutions to these problems include the formulation of a new matrix of data “rearranged” to take these effects into consideration. Nevertheless the “classical” Biplot methods (GH, JK, HJ) reflect both correlations between time intervals and between time series. The Biplot methods provide graphical procedures that offer more information about the behavior of time series than PCA, while making possible the use of more specific quality of representation metrics. In the case of HJ-Biplot, the graphical display obtained has maximum quality of representation for rows and columns.