Self-organized Natural Roads for Predicting Traffic Flow: A Sensitivity Study

TL;DR

This study investigates how self-organized natural roads, based on the Gestalt principle, can predict traffic flow more effectively, revealing a tipping point and surprising insights into network topology and traffic correlation.

Contribution

It introduces a sensitivity analysis of natural roads formation, highlighting the impact of selfish strategies and point-based metrics on traffic prediction accuracy.

Findings

Significant correlation improvement with selfish natural roads.

Existence of a tipping point from segment-based to road-based topology.

Point-based metrics outperform line-based metrics in traffic correlation.

Abstract

In this paper, we extended road-based topological analysis to both nationwide and urban road networks, and concentrated on a sensitivity study with respect to the formation of self-organized natural roads based on the Gestalt principle of good continuity. Both Annual Average Daily Traffic (AADT) and Global Positioning System (GPS) data were used to correlate with a series of ranking metrics including five centrality-based metrics and two PageRank metrics. It was found that there exists a tipping point from segment-based to road-based network topology in terms of correlation between ranking metrics and their traffic. To our big surprise, (1) this correlation is significantly improved if a selfish rather than utopian strategy is adopted in forming the self-organized natural roads, and (2) point-based metrics assigned by summation into individual roads tend to have a much better correlation with traffic flow than line-based metrics. These counter-intuitive surprising findings constitute emergent properties of self-organized natural roads, which are intelligent enough for predicting traffic flow, thus shedding substantial insights into the understanding of road networks and their traffic from the perspective of complex networks. Keywords: topological analysis, traffic flow, phase transition, small world, scale free, tipping point