NYU Abu Dhabi researchers have published one of the most comprehensive reviews on how traffic flows in cities using data from 2.3 billion vehicle observations in 25 different cities.
To create improved systems for traffic modeling and management, a team of researchers, including NYUAD Research Scientist Daniel Bramich and NYUAD Professor of Civil and Urban Engineering and Director of the CITIES Research Center, Monica Menendez, have published a review of previously proposed traffic models using a global data set to compare their accuracy.
The researchers analyzed metrics such as traffic flow, density, and speed to review 50 different models for the flow of traffic and published their findings in a paper titled “Fitting Empirical Fundamental Diagrams of Road Traffic: A Comprehensive Review and Comparison of Models Using an Extensive Data Set” in the journal IEEE Transactions on Intelligent Transportation Systems.
The researchers found that a model with minimal assumptions outperformed the other traffic flow models, regardless of variables such as the road type and the level of congestion.
The winning model is non-parametric, meaning that its number of free parameters can vary as necessary, but without overfitting the data. This represents a departure from traditional parametric models (with fixed numbers of parameters) and opens the door to new approaches to understanding the data with the ultimate goal of improving traffic management and relieving congestion.
“We hope that our review will become the ‘one-stop shop’ for researchers who seek to understand the development of present-day traffic models,” said Bramich and Menendez. “We believe this will serve as an important resource for traffic engineers as they design traffic infrastructure, including traffic light timing and speed limits, for cities around the world.”
This study is the first to complete a fully comprehensive review and comparison of the numerous fundamental traffic models that have been developed over the past 85 years. Specifically, the amount of data this study had access to is approximately 100-1000 times larger than any of the data sets from previous studies, and the number of models studied is approximately ten times larger than previous studies.
With unprecedented access to a large and well-controlled sample of traffic data from cities around the world, the review presented avoids the shortcomings of previous reviews which did not have access to a data set of this scale and did not take all of the top models for traffic flow into account.