New research by engineers at the University of Massachusetts Amherst has found that some of the most important factors that can predict traffic crashes are things like abrupt changes in speed limits and incomplete lane markings. Based on these characteristics, the study then employed machine learning to forecast which roads could be the riskiest.
The study, which was published in the journal Transportation Research Record, involved civil and environmental engineers Jimi Oke, assistant professor, Eleni Christofa, associate professor, and Simos Gerasimidis from the University of Massachusetts Amherst, as well as civil engineers from Egnatia Odos, a publicly owned engineering firm in Greece.
The most significant elements were pavement deterioration (cracks extending across the road and webbed cracking known as “alligator” cracking), road design flaws (such as sudden changes in speed limits or guardrail concerns), and inadequate signage and road markings.
The researchers employed a dataset of 9,300 kilometers of roads spread across 7,000 locations in Greece in order to find these traits. According to Gerasimidis, “Egnatia Odos had the real data from every highway in the country, which is very hard to find.”
Along with Christofa, Oke is a faculty member at the UMass Transportation Center, and he believes the discoveries might have far-reaching implications.
According to him, “the problem itself is applicable globally—not just to Greece, but to the United States.” Although variations in road layouts could affect the ranking of factors, he believes that the features themselves would be significant regardless of location due to their intuitive nature. “There’s no reason to suppose that the indicators wouldn’t be generalizable to the U.S.; the indicators themselves are universal forms of observations.” He adds that this strategy can as easily applied to fresh data from different sources.
A significant step toward improving safety results for all is that it makes good use of decades’ worth of road data: “We have all these measures that we can use to predict the crash risk on our roads,” he states. This work has numerous potential applications in the future. To begin with, it will aid in focusing future research efforts on the most crucial aspects to examine. We had about sixty strange indicators. However, at this point, we can only concentrate all of our resources on acquiring the necessary ones,” explains Oke. “One could delve deeper to understand how a certain feature actually could contribute to crashes,” the author suggests, and then take measurements to determine whether resolving the problem would actually result in fewer instances happening.
Additionally, he sees this as a way to train AI to monitor road conditions in real time. He suggests that as a first step towards an automated monitoring system, “you could train models that can identify these features from images and then predict the crash risk, and also provide recommendations on what we should fix.”
According to Gerasimidis, this is an intriguing use of AI in the real world. He states, “We are doing a big initiative here with specific engineering outcomes.” “This AI study was conducted with the intention of showing [Greek] officials what we are capable of. It is quite challenging to employ AI and produce results that are concrete enough to be put into practice, and this study appears to be one of those. It is now the responsibility of Greek authorities to apply these additional resources to lessen the severe issue of deadly car crashes. We really hope that our research will help to solve this issue.
He goes on, “This work might act as a blueprint for future academic-engineer collaborations on other topics.” “When examining societal issues, the combination of quantitative techniques and actual facts is incredibly potent.”
