By Saumik Masud, Traffic Engineer

Growing up, both of my parents were doctors. From an early age, I saw how meaningful it is to save lives and improve people’s well-being. I have always wanted to apply my own knowledge and skills in transportation safety in a similar way, and this research gave me an opportunity to do just that.

How it all started

Safety analysis plays a huge part in Safe Streets for All (SS4A) projects. Currently, data is effective at identifying roadway hotspots and contributing factors for crashes, but it often leads to generic, unsuccessful countermeasures where similar treatments are applied everywhere. In these situations, one size certainly doesn’t fit all.

As I began analyzing crash data for my first two SS4A projects in El Dorado, Kansas, and Eureka, Kansas, I discovered that while multiple factors contribute to crashes—including driver behavior, roadway design, traffic conditions, vehicles, and the surrounding environment—the available data didn’t always capture the full picture. Additionally, all situations shouldn’t be treated the same. Doing so limits the ability to prioritize solutions that could possibly lead to successful crash reduction strategies.

As I analyzed the crash data for each community, my interest was piqued, and I began exploring new ideas that reached beyond the scope of my daily responsibilities. Was there a way to explore crash patterns to gather more granular data that could be used to prioritize investments and result in specific targeted safety strategies?

As I began to work through that idea, I began to see a consistent gap: many communities still rely on reactive safety improvements after crashes occur, rather than building proactive, data-driven systems that prevent the crashes in the first place.

This led me to focus on developing a more long-term, systemic approach to safety planning—one that integrates crash data analysis, network-wide risk identification, and prioritization of high-impact interventions before incidents happen. I wanted to better understand recurring crash patterns and be able to help agencies select targeted safety improvements where they are most needed.

Research in the real world

Using the El Dorado and Eureka projects as a proving ground, I began to look at all crash factors, not just a single factor. Crashes were weighted by severity—the most serious outcomes had the greatest influence. Fatal and disabling-injury crashes received 15 points, non-disabling injuries received five, possible injuries received two, and property-damage-only crashes received a single point.

The points from nearby crashes were then combined at intersections and along roadway segments to calculate a location-based risk score. These scores were used to rank locations and create a High-Risk Network.

Next, machine learning and clustering algorithms were used to group crashes with similar characteristics together. To understand the relationships of all the factors, a method called the Random Forest model was used on crash data. This method works like a large group of “decision-makers” that each look at the same crash information in different ways and then combine their results to find patterns. Then, a second tool—SHAP analysis—was used to explain which factors were most important in influencing severe crashes. (Notably, machine learning can easily identify complex relationships that may be difficult to recognize through traditional analysis.)

The proportion of each crash pattern at every high-risk location was analyzed to identify the most common safety concerns and guide targeted safety improvements to address those concerns.

At each high-risk location, crashes were grouped by pattern, and the percentage belonging to each pattern was calculated. For example, if 60 percent of crashes are related to speeding, 25 percent involve intersection conflicts, and 15 percent occur at night, speed-related improvements would receive the highest priority at that location. In this way, the proportion of each crash pattern helps determine which safety strategies are most relevant.

The same approach can be applied across the entire roadway network. If similar crash patterns appear at several locations, broader improvements—such as speed management, lighting, enforcement, or public education—can be implemented at multiple sites.

El Dorado case study

Using this new technique, the El Dorado’s project crash analysis showed that the factors most strongly linked to severe crashes were driver distraction, speeding, failure to yield, and impairment. Other important contributing conditions included intersection conflicts, poor lighting, limited sight distance, and inadequate pedestrian facilities.

These results were then grouped together using a method called clustering, which sorts crashes into similar types based on shared characteristics. This helped identify four main crash patterns in the community and laid the foundation for targeted safety strategies.

Crash pattern	Safety strategies
Multi-vehicle and intersection conflicts (crashes at intersections or involving several vehicles)	– Improve intersection visibility, signing, and pavement markings – Evaluate turn lanes and traffic-control upgrades – Apply access management – Consider roundabouts or signals where warranted
Speeding, aggressive driving, and lane-related conflicts	– Apply speed-management and traffic-calming measures – Enhance lane and edge-line markings – Install appropriate warning signs and rumble strips – Conduct targeted enforcement and education
Rear-end crashes and parking-related conflicts	– Improve access management and driveway spacing – Evaluate turn lanes – Restrict parking near intersections and driveways where visibility is limited – Improve parking-area circulation and delineation
Nighttime crashes involving impaired driving	– Improve roadway and intersection lighting – Enhance the retroreflectivity of signs and pavement markings – Conduct targeted impaired-driving enforcement and education – Support safe-ride programs

Eureka case study

The study proved successful in Eureka, too. That community’s project crash analysis showed that six crash patterns most strongly influenced severe crashes, leading to specific strategies.

Note that locations in close proximity can sometime have differing root causes. This evidence-based prioritization helps identify projects for both short- and long-term implementation strategies.

Looking toward the future

Ultimately, now that this innovative strategy has been applied successfully in real-world projects, its myriad benefits can apply to agencies of any size in any location.

• It outlines a repeatable, data-driven framework that can be used to identify high-risk locations, understand the dominant crash patterns at each location, and select more targeted safety strategies.
• The results provide a foundation for long-term safety planning by helping agencies focus improvements where they are expected to provide the greatest benefit.
• The framework can be updated as new crash and roadway data becomes available, allowing agencies to track changing safety conditions and evaluate whether implemented improvements are working.
• The same process can be applied in other communities using their local crash, roadway, traffic, and environmental data. However, models and crash patterns should be recalibrated for each community because the factors contributing to crashes may differ by location.
  

Interested in learning more?

This research has been presented to a broad audience at regional, national, and international conferences, and I’d like to share it with you, too. If you’re interested in learning more about the process or how we can apply this methodology to your community, please reach out to me at smasud@jeo.com. I’d love to get the conversation started.