Fast and Scalable Traffic Volume Estimation using Modified betweenness Centrality

Many applications in transport systems analysis and planning-from travel time and accessibility calculations to environmental impact and noise assessments-make use of traffic volume and aggregate traffic flow estimates. Travel demand models can generate such estimates, but they often require extensive input data, long computational times, and are complex to validate. As a potentially less accurate but faster alternative, network science concepts like centrality measures have emerged. By default, however, such network topology measures cannot yield realistic traffic volume estimates, as they ignore demand-side characteristics of transportation networks. This study explores combining a network topology measure, edge betweenness centrality, with distance-decay-based node sampling to estimate annual average daily traffic (AADT) for cars on each road segment in a network serving over 3 million people. Further improvements to these AADT estimates are tested by refining the node sampling with location-specific data and by employing a simple capacity constraint mechanism. The final model achieves a Pearson correlation coefficient of 0.90 and a mean absolute error (MAE) of 4,100 vehicles/day compared to traffic counter data, approaching accuracies of full-stack travel demand models. Even a basic approach, requiring little data beyond a simple road network representation and a single distance-decay function, achieves 85-90% of this accuracy. These results suggest that fast, simple, and scalable traffic estimation methods using modified network centrality measures can be viable when reduced data and computational requirements are prioritized over maximum accuracy.