Machine learning used to evaluate 1,773 bus stops across Massachusetts to determine risk factors and inform safety countermeasures
AMHERST, Mass. — With pedestrian fatalities—particularly in public transit areas—continuing to rise across the country, researchers at the University of Massachusetts Amherst have determined the top risk factors of pedestrian-vehicle crashes at bus stops to recommend potential solutions. While their comprehensive analysis of bus stops focused on Massachusetts, the researchers are excited about the generalizability of the findings and application to other locations.
"Despite significant advances in road design over the past decade to enhance safety, we are still seeing safety issues that are disproportionately impacting pedestrians," says Tolu Oke, lead study author and a Ph.D. candidate in transportation engineering at UMass Amherst.
Since the pandemic, bus ridership across the U.S. has recovered more significantly than other forms of public transit, recovering 86% of pre-pandemic ridership levels by 2025 and previous research shows that 46% of crashes in Massachusetts involving pedestrians happen near bus stops.
"Is this the chicken or the egg?" asks Michael Knodler , the William W. Boyer Endowed Professor of Engineering in the Riccio College of Engineering at UMass Amherst, director of the UMass Transportation Center, and one of the authors of the paper. "Are the crashes happening there because that's where the pedestrians are, or is there something about the particular aspects of bus stop design or placement contributing to the pedestrian risk?"
Oke and her team used machine learning-based factor analysis to examine 1,773 such stops, ultimately determining 13 crash-relevant characteristics that cluster into 13 distinct bus stop types ranging from stops on remote rural roads to densely packed urban streets.
The 13 characteristics can be organized in three categories: First are the immediate-environment factors. This includes bus stop infrastructure, road condition, road width, crossing difficulty or crossing safety. Second are network integration and connectivity factors: Interstate highways, principal or minor arterials, or major or minor connectors. And finally, there are sociodemographic and land-use context, describing the characteristics of the community and area the stop serves such as core density, social advantage or social equity. Unique clusters of these factors make up the 13 distinct bus stop types.
The two bus stop types with the highest prevalence of crashes across all the clusters are in "mixed-use arterial corridors" and "dense urban cores" areas.
While slightly different in their makeup, both are complex environments characterized by high traffic volumes, multiple modes of transportation and a mix of properties, such as apartments, stores or parks.
"All the activity in these areas just provides more opportunity for conflict," says Oke.
Dense urban cores like Broadway and City Hall Avenue in Chelsea or State Street and School Street in Springfield have more pedestrians, while mixed-use arterial corridors like Harvard and Beacon streets in Brookline or Mill and Locus streets in Springfield tend to feature wider roads lined with businesses and apartments.
Based on the mix of characteristics, the researchers proposed recommendations specific to each bus stop type. "We can't apply the same safety solution everywhere," says Oke. "Our research shows that different areas have different crash problems. In one area, the solution might be the addition of a crosswalk, while on a dense, multimodal corridor, we may need protective features to separate and calm the complex interaction of pedestrians, cyclists, cars and buses."
Some countermeasures aim at slowing drivers down, not because it will reduce the number of crashes, but because it will make those that happen less deadly.
"The overwhelming countermeasure is speed," Knodler adds. "Whenever you have these potential conflicts, managing the speeds across the different modes is the key. If a pedestrian is hit by a driver in a car going over 40 miles an hour, there's an 80% probability that that pedestrian is seriously injured or killed. If that speed is 20 miles an hour, it's less than 20% chance that that pedestrian is seriously injured or killed."
The full list of the UMass Amherst study's characteristics, bus stop types descriptions and proposed countermeasures can be found at Data Science for Transportation .
