AI Might Know Where You’re Going Before You Do Thanks to a New Human Movement Prediction Model
Artificial intelligence is getting better at understanding how humans behave, and a new research project suggests it may soon predict where people are headed even before they decide themselves. Researchers at Northeastern University have developed a system that uses large language models (LLMs)—the same type of AI behind chatbots and text generators—to forecast human movement patterns with surprising accuracy.
The research focuses on a tool called RHYTHM, short for Reasoning with Hierarchical Temporal Tokenization for Human Mobility. At its core, RHYTHM applies language-model-style reasoning to location data, allowing AI to recognize patterns in how people move throughout the day, week, and even month. The goal is not just prediction for prediction’s sake, but practical improvements in areas like transportation planning, traffic management, and emergency response.
What RHYTHM Is and Why It Matters
Predicting human movement has always been a difficult problem. People appear to act randomly, especially at the individual level. One day someone goes straight home after work; another day they stop at the grocery store, meet friends, or head to the gym. Traditional mobility models struggle to handle this apparent randomness.
RHYTHM approaches the problem differently. Instead of treating human movement as chaotic noise, it assumes that most people follow underlying rhythms. These rhythms might not be obvious at first glance, but when viewed over time, they reveal repeating patterns—commutes on weekdays, errands on weekends, or recurring weekly activities.
By using open-source mobility data and the contextual reasoning abilities of LLMs, the researchers built a system that doesn’t merely copy past trajectories. Instead, RHYTHM reasons about why people move the way they do under certain conditions and uses that understanding to predict future locations.
How RHYTHM Predicts Human Movement
One of RHYTHM’s key innovations is how it processes movement data. Traditional models often analyze entire movement trajectories as long, continuous sequences. This can be computationally expensive and inefficient, especially when patterns repeat over time.
RHYTHM breaks trajectories into temporal tokens. These tokens represent chunks of movement tied to specific time intervals, such as parts of a day or recurring time slots across multiple days. This method allows the model to better capture periodic behavior, such as daily routines or weekly schedules.
The system also uses a hierarchical structure, meaning it can understand both short-term movements (where someone might go in the next 30 minutes) and longer-term patterns (where they are likely to be over the next 24 hours or even several days). This layered understanding helps the AI handle both regular routines and occasional deviations.
Accuracy Gains and Performance Advantages
According to the research findings, RHYTHM performs better than existing state-of-the-art mobility prediction models. Overall, it achieves about 2.4% higher accuracy compared to similar systems. While that number may seem small, it is significant in a field where even marginal improvements are difficult to achieve.
The performance gap becomes even more notable during irregular periods, such as weekends or non-workdays. During these times, RHYTHM is roughly 5% more accurate than competing models. These irregular periods are usually the hardest to predict because people deviate from their usual routines.
Another important advantage is efficiency. Training large language models can be extremely time-consuming, but RHYTHM requires significantly less training time. This makes it more practical for real-world applications where models need to be updated frequently with new data.
Testing the Model in Real Scenarios
To evaluate RHYTHM, the researchers trained it using seven days of human movement data. After that, they tested how well it could predict future movement.
The results showed that RHYTHM could reliably forecast where people would be:
- The next day
- Over the next few days
- Up to one week into the future
While the model can theoretically predict further ahead, the researchers found that errors tend to accumulate over longer time horizons. For this reason, RHYTHM is especially well-suited for short- to medium-term forecasting, which happens to align with many real-world needs.
Potential Applications Beyond Daily Life
The most immediate applications of RHYTHM are in transportation and traffic planning. By understanding how people move through cities, planners could optimize public transit schedules, reduce congestion, and design infrastructure that better matches actual usage patterns.
However, the researchers are particularly interested in how RHYTHM could help during extreme events. Natural disasters, major accidents, or large-scale emergencies often disrupt normal routines, making human movement even harder to predict. Knowing where people are likely to go in the next few hours could be critical for evacuations, emergency services, and disaster response coordination.
Because RHYTHM can handle uncertainty and rare events better than older models, it may be able to simulate scenarios that are difficult to observe directly but still important to plan for.
Why Large Language Models Are Useful Here
It might seem odd to use language models for predicting movement, but there is a good reason behind this choice. LLMs excel at recognizing patterns, context, and structure in complex data. Human mobility, much like language, follows implicit rules and rhythms.
By borrowing techniques from natural language processing, the researchers enabled RHYTHM to reason about movement data in a more flexible way. Instead of rigid mathematical assumptions, the model learns nuanced relationships between time, location, and behavior.
This reflects a broader trend in AI research: applying language-model reasoning to non-language problems, including biology, finance, and now human mobility.
Broader Context: The Future of Mobility Prediction
Human mobility prediction is becoming increasingly important as cities grow and systems become more interconnected. Smartphones, GPS devices, and smart infrastructure generate massive amounts of location data, creating both opportunities and challenges.
While tools like RHYTHM can improve efficiency and safety, they also raise important questions about privacy and data governance. The researchers emphasize that their work relies on open and anonymized datasets, but wider adoption will require careful regulation and ethical oversight.
At the same time, the ability to predict movement patterns could lead to smarter cities, better emergency preparedness, and more responsive public services.
What Comes Next for RHYTHM
The research team plans to further refine RHYTHM, especially for use during rare and high-impact events. Future work may involve integrating additional contextual data, such as weather conditions or real-time infrastructure disruptions, to make predictions even more robust.
As AI systems continue to expand beyond text and images into real-world behavior, tools like RHYTHM offer a glimpse into how predictive intelligence might shape everyday life in the near future.
Research paper: https://arxiv.org/abs/2509.23115
