Citation

Flynn DFB, Gilmore MM, Dolan JP, Teicher P, Sudderth EA. Transp. Res. Rec. 2022; 2676(8): 267-278.

Copyright

(Copyright © 2022, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.1177/03611981221083305

PMID

unavailable

Abstract

Crowdsourced mobile applications such as Waze can provide real-time and historical data about roadway conditions, when and where users are active. In a previous study, we demonstrated that statewide crash models based on integrated Waze, traffic volume, census, and weather data give reliable hourly estimates of police-reportable crashes in 1-mi area grids at 1-h timescales. Here, we extend our analytical methods to test an application of Waze traffic alerts to a crash prediction model used to guide law enforcement resource allocation. The Crash Reduction Analyzing Statistical History (CRASH) model is used by the Tennessee Highway Patrol (THP) to prioritize patrol locations. The model combines historical data such as fatal crashes with current data, including weather forecasts and scheduled special events, to identify areas with a high likelihood of crashes. To more accurately target locations and times with a high crash propensity, we assessed the potential for Waze alerts to improve the temporal and spatial resolution of the CRASH model. We found that with Waze data, we increased the spatial resolution of crash estimates from 42 to 1 mi2 and the temporal resolution from 4- to 1-h time windows, while improving accuracy. The model provides a high-resolution option for the allocation of patrols, which will help THP to optimize the allocation of troopers to the highest-risk locations. Beyond the current implementation in Tennessee, the model's incorporation of crowdsourced data has shown potential for similar types of data-driven safety approaches elsewhere.

Language: en

Keywords

artificial intelligence and advanced computing applications; crash analysis; crash prediction models; data and data science; machine learning (artificial intelligence); safety; safety performance and analysis; supervised learning