Exploring Statistical Backbone Filtering Techniques in the Air Transportation Network

The dense nature of transportation networks expands the challenge of their visualization and processing. Several statistical backbone extraction techniques are proposed to reduce their size while keeping essential information. Here, we perform a comparative evaluation of seven prominent statistical backbone extraction techniques in the USA weighted air transportation network. One can classify the airports into hubs, spokes, and focus airports based on the business models used by the airlines. We compare the extracted backbones using various performance measures. We consider the number of components, sizes, the fraction of airport type, edge type, and weights preserved by each method. Results show that the Enhanced Configuration Model (ECM) Filter tends to preserve edges between spoke airports uncovering the infrastructure connecting the regional spoke airports. In contrast, the alternative filters (Disparity, Polya Urn, Marginal Likelihood, Noise Corrected, Global Statistical Significance (GLOSS), Locally Adaptive Network Sparsification (LANS)) highlight edges between the hub and spoke, focus and spoke, and spoke and spoke airports revealing more of the hub and spoke foundation used by airlines. Moreover, the Disparity Filter, Marginal Likelihood Filter, and Noise Corrected Filter preserve the highest proportion of weights while Polya Urn Filter and ECM Filter keep the lowest. The GLOSS and LANS Filters maintain a moderate fraction of weights between the two extremes.