Unveiling Roadway Network Safety: Application of Statistical Physics to Crowdsourced Velocity Data

The incessant rise in vehicle-caused fatality rates worldwide motivates us to investigate the near-miss collision risk in traffic flow using attributes of statistical physics. We find that the statistics of accident precursors can be mapped onto a phase diagram that elevates observables, such as traffic density and velocity fluctuations, to predictive metrics of accident risk. To this end, we derive the near-miss collision risk from the velocity state transition matrix, considering the excessive deceleration of particles in steady-state traffic flow. We probe our model against actual collision data using both simulation-based and crowdsourced vehicle velocity data. We show that an intrinsic near-miss collision risk exists, is confined to congested flow, and predicts the highest likelihood of actual collisions. This risk decreases with increasing randomness in driver behavior—a feature it shares with many other many-body systems with long-range correlations triggered by randomness, e.g., random agents moderating price fluctuations in financial markets. We further advance our discussion by applying the methodology to large-scale crowdsourced velocity data across various states in the United States. Specifically, focusing on a network's ability to sustain functionality and connectivity amid potential disruptions, we examine the reliability and resiliency of roadway networks at the state scale by upscaling path-related attributes.



We demonstrate that our statistical-physics-inspired approach not only allows for identifying high-risk road segments or zones but is also instrumental in assessing the broader safety contexts, such as the reliability and robustness of networks. More importantly, the availability of a predictive phase map of accident precursors and its scalability to large velocity datasets is expected to enhance global road safety, including early warning systems for user-operated and self-driving vehicles.