How many people will travel during a given week between two specific cities? Answering this question is important for many reasons, for example, to design efficient public transport infrastructures, or, as was the case during the COVID-19 pandemic, to understand how mobility patterns could be used to predict the spread and evolution of the virus. Now, a new mathematical model developed by the URV's SeesLab research group , together with researchers from Northeastern University and the University of Pennsylvania in the United States, has made it possible to predict human mobility with high precision and in a simpler and more efficient way than the systems currently in use. The scientific journal Nature Communications has published the results of the study, which provides a valuable new tool for understanding how people move in different contexts.
Models of human mobility have existed for decades. Since the middle of the 20th century, so-called "gravitational models" have been used to understand and predict human mobility. These systems are inspired by Newton's law of gravitation and, in order to give their results, take into account two fundamental parameters: the size of the population of the two cities and the distance between them. These models assume that larger populations attract more movement, while larger distances act as a disincentive. Gravitational models have been used in transport planning, migration studies and epidemiology because they make it possible to understand the results very simply and to predict spatial interactions and flow patterns. However, this simplicity means that these models are not extremely accurate and can only give approximate predictions of mobility flows.With the advent of artificial intelligence, in recent years the research community has begun to develop much more accurate mobility models based on machine learning. Unlike the original gravity models, which predict flows only from population and distance, these new models use many more variables besides origin and destination, such as the density of restaurants and schools or road connectivity. Although their predictions are much more reliable, unlike gravity models, the results are very difficult to interpret and do not offer a clear view of the mechanisms that explain people's mobility decisions.
Now, the URV research team has managed to combine the best of each system: the accuracy of the machine learning models and the simplicity of the gravitational systems. Based on an algorithm they call the "scientific robot" , they have developed a new innovative mathematical model that equals and even improves the accuracy of the machine learning models and, moreover, is as simple and easy to interpret as the gravitational model. "With this new algorithm we can identify the most plausible models for explaining the data observed, in our case, mobility flows", explains Marta Sales-Pardo, researcher at the SeesLab research group. The method combines machine learning techniques, statistical physics and Bayesian statistics to efficiently balance the complexity of the model and its accuracy. "We have developed a very powerful tool for scientific discovery and data-driven modelling," says Roger Guimerà, ICREA research professor in the same group.
The pre-doctoral researcher Oriol Cabanas, who also took part in the study, pointed out that the model can also be extrapolated to other geographical areas. "As it only uses population and distance variables, only minimal adjustments to the parameters are needed to extrapolate its predictions to other geographical areas". Thus, this new approach can be used to analyse displacements both in large cities and in less urbanised areas without having to create a new algorithm, as would be the case with machine learning models, due to their complexity.
Understanding human mobility is fundamental in many areas. For example, in urban planning and transport, the model can help road infrastructure and public transport services be planned more efficiently by optimising resources and reducing congestion. It is also useful and necessary in the field of public health because it can be used to model the spread of infectious diseases by understanding how people move and how viruses and other pathogens can be transmitted from one area to another, and by designing containment strategies in the event of a pandemic.
Furthermore, the model's ability to predict human mobility also has implications for sustainability, as it can help to better manage energy consumption and reduce the greenhouse gas emissions associated with transport.
The research by the SeesLab research group does not stop there. In fact, they have already begun to test the model with other variables in addition to population and distance, such as road connectivity, and the results suggest that it can produce an even more accurate picture of mobility.
