Predictive models could reduce car crashes, Mineta research finds

© Shutterstock

Researchers at the Mineta Transportation Institute have developed accurate predictive models for traffic congestion and road accidents using statistical techniques and machine learning algorithms.

Annually, more than 8,000 people die in crashes on U.S. roads, and 2.35 million are injured or disabled, according to a 2020 Association for Safe International Road Travel report.

The research team trained the predictive models using four machine learning algorithms and then compared the models. Researchers discovered all methods have similar accuracy using the original imbalanced data.

When researchers examined the data, they discovered the highest number of accidents occurred between 4 pm to 6 pm, and weekdays had the highest number of accidents. October, November, and December have more accidents than other months.

“This is a highly complex issue,” Dr. Hongrui Liu, lead investigator, said. Fortunately, we have a large dataset – 4.2 million records of car accidents from 2016 to 2020 – to help us train these models and take steps toward improving safety and mobility.”

Results can be used to improve road safety and understand traffic congestion. The predictive models can be used to build a smart transportation system when paired with the real-time visibility of environmental conditions offered by advanced information technologies.