On the Correlation and Predictability of Topological Measures in Transportation Networks

The computation of topological measures in large-scale complex networks, such as those found in transportation systems, is often a resource-intensive process. These measures, however, are critical for a comprehensive understanding of network structures and for optimizing their design. A key challenge lies in selecting the appropriate metrics that encapsulate the essential information of the network, thereby reducing the computational burden. Traditional methods involve identifying correlations between various topological measures to infer missing data. In this paper, we introduce an enhanced analytical framework comprising three stages aimed at selecting a subset of metrics to efficiently summarize network characteristics and predict measures that are costly to compute. The methodology involves: a correlation analysis of topological metrics; a principal component analysis to reduce dimensionality and highlight the essential features; and the application of SHAP and recursive feature elimination to assess the predictive significance of each metric. We demonstrate the utility of this approach using metro and road networks from 46 cities in the EU/EEA region, yielding promising results in identifying relationships between metrics and predicting missing data.