A Quantitative Analysis-based Algorithm for Optimal Data Signature Construction of Traffic Data Sets

Abstract

In this paper, a new set of m-dimensional Power Spectrum-based data signatures is derived to obtain better Vector Fusion 2-dimensional visualizations of a time series and periodic n-dimensional traffic data set as compared with visualizations produced from using the entire set of n-dimensional Power Spectrum representations in literature, where m « n. We were able to ascertain that 4-dimensional data signatures provide empirically optimal representations with respect to the data set used. We have achieved ≈ 97.6% reduction in terms of data representation of the original nD data set with the signatures. We propose an algorithm that determines how good the selected set of m-dimensional signatures represents the n-dimensional data set in 2 dimensions in quantitative terms. We use the Vector Fusion visualization algorithm in transforming each signature from m dimensions into 2 dimensions. An improved set of qualitative criterion is drawn to measure the goodness of the 2-dimensional data signature-based visual representation of the original n-dimensional data set. Finally, we provide empirical testing, discuss the results, and conclude the contributions of the proposed methods.