Scientists at the University of Sharjah have developed a new machine learning model capable of predicting whether a driver is likely to be involved in an accident before getting behind the wheel.

Road accidents are frequently linked to human error, yet traditional driver screening methods, particularly in taxi and commercial transport sectors, tend to rely heavily on experience and background checks. According to the researchers, these criteria often fall short of predicting who may pose a higher risk on the road.

Their study, published in the journal Engineering Applications of Artificial Intelligence, explores a fundamental question: How can we identify risky drivers before they take to the road?

To answer it, the researchers developed a data-driven assessment framework combining psychological profiling, physiological monitoring, and simulated driving performance.

Participants began by completing a structured questionnaire measuring personality traits such as sensation seeking and conscientiousness. They then drove in a highly realistic simulator designed to replicate urban traffic conditions in Dubai, a cosmopolitan city known for its chronic traffic congestion and road network that handles more than 3.5 million vehicles daily.

“During the session, we recorded heart rate and detailed eye-movement indicators, including blink rate and gaze deviation,” said Dr. Malek Masmoudi, the study’s lead author and an associate professor of industrial engineering. “Using machine learning models, we analyzed this integrated dataset to classify drivers as low-risk or high-risk based on objective outcomes such as accidents and traffic violations recorded during the simulation.”

Distinguishing between safe and risky drivers

The significance of the work lies in its preventive and developmental approach. Rather than responding to accidents after they occur, the proposed framework allows transportation companies to evaluate risk tendencies before they are deployed.

Beyond supporting safer, more informed hiring decisions, the model can also function as a targeted training tool, helping drivers recognize their risk patterns and enhance their attention and self-regulation. This contributes not only to safer roads but also to more structured and evidence-based driver development programs.

“The analysis revealed that gaze distraction, sensation seeking, conscientiousness, and gender are the best predictors of driving behavior,” the scientists note. “The findings suggest that our model can serve as a valuable decision-support tool for taxi companies and transportation agencies aiming to enhance driver selection processes by identifying drivers with lower accident risks.”

According to the authors, the results validate the practical utility of the proposed classification framework for distinguishing between safe and risky drivers. “The findings provide valuable insights for taxi companies and transportation agencies aiming to enhance driver selection processes by identifying drivers with lower accident risks by focusing on highly important factors identified in the study,” they write.

Dr. Masmoudi emphasized that road safety should not begin after an accident; it should begin before a driver is hired. “The safest accident is the one that never happens. That’s why safety must start before a driver ever touches the steering wheel,” he said. “In the age of Industry 4.0, we have the tools to predict risk instead of just reacting to crashes. The question is no longer, "How can we measure it?" It’s: "Why aren't we using it?”

Disciplined and responsible individuals drive cautiously

The findings show that safe driving is not shaped only by technical skills, but it is also strongly influenced by personality traits and patterns of visual attention. For instance, drivers whose gaze frequently strays from the road are significantly more likely to be involved in simulated accidents.

Dr. Masmoudi explained, “Individuals who are naturally disciplined and responsible tend to drive more cautiously. Those who score high in sensation seeking, meaning they are drawn to excitement and risk, are more likely to exhibit unsafe driving behaviors.”

In simple terms, certain measurable personality traits and attention habits consistently indicate a higher risk of accidents. When these factors are analyzed together using machine learning, they allow taxi or transport companies to predict risk with a high level of accuracy.

Beyond screening, the study’s insights can also inform targeted training programs. Drivers identified as higher risk can receive specialized coaching in attention control, stress management, and safe decision-making. For taxi companies and fleet operators, this approach could lead to fewer accidents, reduced insurance and repair costs, improved passenger safety, stronger public trust, and an overall enhancement of company reputation.

“Risky driving is not random behavior; it reflects measurable patterns in attention and personality,” stressed Imad Alsyouf, professor of industrial engineering at the University of Sharjah and a co-author. “By combining psychology, physiology, and machine learning, we move from intuition-based recruitment to evidence-based safety decisions.”

Real-world implications

Prof. Alsyouf emphasized that although their model successfully identifies high-risk drivers, “artificial intelligence should not replace human judgment; it should strengthen it with objective data. Our goal is not only to filter risk but also to help drivers understand and improve their own safety profile.”

Prof. Alsyouf explained that the research was designed with real-world implementation in mind. “Its most immediate application is in taxi and commercial fleet recruitment. Instead of relying solely on driving history or interviews, companies can incorporate a short simulator-based assessment combined with psychological screening and physiological monitoring.”

Importantly, the framework extends well beyond hiring. Dr. Masmoudi pointed out that the model could serve as a structured training and evaluation tool, enabling companies to design personalized improvement programs for drivers.

“The key advantage is that the system operates before deployment,” he added, “shifting safety management from a reactive model, responding after incidents occur, to a preventive and developmental one, where risks are identified early and addressed proactively.”