Air Transportation Network Backbone Extraction: A Comparative Analysis of Structural Filtering Techniques

In the age of advanced data collection tools, large-scale network analysis presents significant visualization and data processing challenges. Backbone-extracting techniques have emerged as crucial tools to tackle this challenge. They aim to reduce network size while preserving essential characteristics. One can distinguish two primary approaches: structural methods, which prioritize nodes and edges based on their topological properties, and statistical methods, which focus on their statistical relevance within the network data. This study investigates eight popular structural methods in an air transportation case study. Correlation analysis reveals that shortest path-based methods yield similar backbones, while Doubly Stochastic and H-backbone methods do not correlate with their alternatives. Interestingly, H-backbone retains high-weight edges, and High Salience Skeleton and Doubly Stochastic backbones capture diverse weight scales. We evaluate the original network information loss using the backbone’s edge, node, and weight fraction. Doubly Stochastic and H-backbone methods keep substantially more edges compared to others. H-backbone, High Salience Skeleton, and Doubly Stochastic uncovered backbones fail to retain all nodes. Connectivity and transitivity comparisons indicate Primary Linkage Analysis, High Salience Skeleton methods disrupt the connectivity, and the Doubly Stochastic preserves the transitivity. This study sheds light on the strengths and weaknesses of these techniques, facilitating their application in real-world scenarios and inspiring future research directions in network analysis.