How prediction analysis can help further cut traffic deaths

The National Highway Traffic Safety Administration estimates 39,345 people died in traffic accidents in the U.S. in 2024. While that figure is staggering, it actually marks a decrease of nearly 4% compared to the 40,901 fatalities reported in 2023, and represents the first time since 2020 that the number of U.S. traffic deaths fell below 40,000.

While the decline in traffic fatalities can be attributed to numerous factors — from advancements in vehicle safety technology to stricter enforcement of traffic laws to changes in driver behavior — a portion of the credit must also be given to auto accident prediction analysis, a tool within the data-driven safety analysis framework that uses data to predict the likelihood of crashes and the number of crashes expected at a particular location or roadway segment, helping government agencies to prioritize safety interventions.

Used in some form by numerous state and local governments throughout the U.S., auto accident prediction analysis employs a wide range of methodologies to identify risk factors and aid in implementing preventative measures designed to reduce traffic accidents. These include:

• Statistical models to predict accident occurrence based on factors like traffic flow, weather conditions, and road geometry
• Machine learning algorithms (such as decision trees and artificial neural networks) to analyze large datasets and predict accident likelihood and severity
• Data mining to extract patterns from historic accident data for use in identifying contributing factors and predicting future occurrences
• Scenario analysis to simulate accident-causing conditions and assess their potential impact
• Time-based patterns to predict future accident trends
• Markov Chain models to predict the probability of an accident occurring at a particular time and location

Used separately or in combination, these predictive methodologies can go a long way toward helping traffic planners and engineers to identify accident hotspots, informing future traffic management strategies while reducing accident severity. The same technologies can be used to improve accident investigations, evaluate the effectiveness of safety policies, and ultimately help state and local governments in prioritizing roadway safety initiatives in which to allocate limited funding.

That funding factor goes well beyond the standard concerns about government budget limitations and an inability to tax citizens beyond a certain threshold. Aside from the obvious loss of life costs, the estimated economic cost of traffic accidents in the U.S. currently exceeds $417 billion annually. That translates into an annual crash tax of nearly $1,268 per U.S. resident, according to a recent report by Advocates for Highway & Auto Safety.  

Unfortunately, while auto accident prediction analysis offers the potential to reduce the quantity and severity of traffic accidents, it has its limitations. Because accidents can be influenced by numerous factors, including the weather, the physical dimensions of the roadway, the actions of other drivers and more, it is difficult to create prediction models that are completely reliable. 

Moreover, accurate models require large, high-quality datasets containing information about crashes, drivers, vehicles, weather conditions, and other potential influencing factors. Such large, detailed datasets are often unavailable and, even when they are, may yield false positives.

Beyond less than reliable datasets, the cost of developing and implementing accurate analysis methodologies can be too high for many jurisdictions. Specialized expertise is also needed to interpret the data yielded for auto accident prediction analysis, which again could present a budgetary constraint for many government agencies. State and local governments also have to balance the positive that can be achieved by employing such methodologies with the demand to dedicate funding to effective accident response and investigation.

There is some good news, however, in the form of new research designed to improve the accuracy and reliability of accident prediction technology. Enabling datasets to account for a much wider range of factors, such as construction, population density, real-time traffic conditions, various road surfaces and more could lead to the development of more robust accident prediction methodologies that can then be applied across varied geographic locales and traffic scenarios.

Significant advances in machine learning and artificial intelligence also present an opportunity to capture underlying data relationships and integrate more sophisticated predictive models. Optimizing such inputs, in turn, can lead to more precise, real-time predictions capable of navigating the complexity inherent in traffic data.    

Similarly, the application of artificial neural networks to analyze crash sites has been limited to date, despite its success in predictive tools such as driver behavior analysis, vehicle detections, and traffic signal control. The findings of several pilot programs, however, suggest that neural networks models can be more effective in predicting crash frequency than traditional statistical methods.

These developments, coupled with significant advancements in autonomous driving technologies and advanced driver assistance systems, as well as wider use of the current iterations of auto accident prediction analysis by state and local governments, give us hope that the decline in traffic fatalities experienced over the past few years will not only continue, but accelerate, even as traffic volume (and the speed at which those vehicles drive) continues to rise. 

Wes Guckert, PTP, is President & CEO of The Traffic Group, a Service-Disabled Veteran-Owned Small Business (SDVOSB), Maryland-based traffic engineering and transportation planning firm. For more information: www.trafficgroup.com or follow them on LinkedIn.