CDS&E: Point Process Models for Traffic Risk Analysis and Crash Prevention

Traffic and crash databases collected by state and federal departments of transportation contain a wealth of information that can be used to increase highway safety. For example, crash databases can be used to identify locations where crashes frequently occur as well as the underlying factors that contributed to those crashes. Such analysis of crash databases can subsequently lead to identifying countermeasures that can be enacted to decrease the frequency of crashes at high-risk locations.

While many efforts are ongoing to use the information contained in these databases to increase traffic safety, modern traffic datasets contain more information than can be currently extracted using traditional data analysis techniques. The most glaring shortcoming of common statistical techniques for crash data is that such techniques focus only on small segments of the road (e.g. intersections) rather than analyzing the entire roadway network simultaneously.

In this project, the researchers are developing statistical methodology that will analyze an entire roadway network to capture important relationships between roadway features that may lead to an increase in crashes. Ultimately, the goal of this project is to analyze traffic network data so as to identify ways to create a safer roadway network for all travelers.

Beyond research activities, mentoring activities associated with this project include student mentoring on advanced topics in data science and traffic safety engineering. Educational activities will include STEM career presentations to high school students as well as the development of a novel interdisciplinary research group.

Historically, statistical models for traffic crashes have analyzed aggregated crash counts along with roadway segments, where aggregated data forfeit the use of any within-segment information. Modern crash databases, however, contain data on the exact locations of crashes (referred to as point pattern data) which, if analyzed appropriately, can give richer statistical inferences than aggregated data.

This project seeks to fully utilize the information in modern traffic databases by considering the continuous nature of roadway traffic rather than relying on arbitrarily aggregated count data over roadway segments. Specifically, this project will develop easily implementable and computationally feasible approaches to modeling crash point pattern data to determine where crashes are likely to occur (referred to as hot spot identification) as well as how features of the roadway influence the potential for a crash (referred to as risk factor identification).

Specifically, this project has the following goals related to statistical and civil engineering science: (1) develop piece-wise linear point process models on roadway networks and (2) develop a hierarchical point process approach to model multiple crash types simultaneously.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.