Artificial and collective intelligence for the most accurate prediction of human movements

mobility

Predicting where people will go is more than a scientific challenge—it’s a practical necessity for tackling today’s pressing issues, from urban planning and traffic optimization to epidemic containment.

While Artificial Intelligence (AI) already performs well in tracking routine movements based on historical data and learned patterns, it struggles when behavior becomes irregular. How can we forecast movements that have never happened before?

This question lies at the heart of the study “Mixing Individual and Collective Behaviours to Predict Out-of-Routine Mobility,” recently published in the prestigious journal PNAS and onducted by a team of Fondazione Bruno Kessler‘s researchers in collaboration with the Institute of Science and Information Technology at CNR.

When individual movement data are insufficient or erratic, the model leverages collective behavior—drawing on patterns of how people typically move within a specific urban environment. By doing so, it can make more accurate predictions even in unfamiliar or exceptional circumstances.

The experiments, based on the analysis of hundreds of thousands of anonymized trajectories in three major US cities, showed that this approach is significantly more effective at predicting non-routine travel than traditional AI techniques. This improved accuracy provides critical support for decision-making in fields like city planning, traffic control, and crisis response— be they health-related or environmental.

“Our mobility is easily predictable in most situations. Most of us frequent a small number of places, and many AI models handle this kind of routine prediction quite well,” explains Massimiliano Luca, lead researcher at FBK’s Mobile and Social Computing Lab (MobS), FBK Ambassadorand study’s coordinator. “However, during extraordinary events like the COVID-19 pandemic or natural disasters, traditional models fall short. They fail to predict mobility shifts and thus can’t provide timely, actionable insights to policymakers.”

“We also found that in densely populated urban areas with a high concentration of points of interest, collective behavior becomes a stronger predictor of out-of-routine movements,” adds Sebastiano Bontorin, MobS researcher and lead author of the paper.

The study was carried out by FBK researchers Sebastiano Bontorin, Simone Centellegher, Riccardo Gallotti, Bruno Lepri, and Massimiliano Luca, in collaboration with CNR researcher Luca Pappalardo.