Mobility: using big data to revolutionize theoretical models

These days, data is in abundant supply – and that supply is growing exponentially. This is a godsend for researchers, provided that they know how to listen and let the data speak for itself. And there are undeniable advantages if they do, as illustrated by Marija Nikolic’s thesis on pedestrian-flow modeling, which she recently defended. Her models are driven by large amounts of real data, making them more reliable, robust and detailed and helping us to better understand and more accurately characterize pedestrians’ movement behavior.

Three fundamental variables are used to describe and model pedestrians’ movements: speed (unit of distance per unit of time), density (number of individuals per unit of space at a given moment) and flow rate (number of individuals passing a given point per unit of time). The fundamental relationships between these three variables can be determined theoretically using equilibrium hypotheses. Nikolic, however, takes a different approach in her thesis: her model uses real data to quantify these variables and determine the relationships between them, attaching less importance to arbitrary hypotheses.

The researcher had a goldmine at her fingertips: she used the data collected at Lausanne’s train station by sensors that closely – and anonymously – track pedestrians’ movements. “It's the first time so much real data has been available, without the data being distorted by the experiment itself,” says Nikolic. And in order to identify the busiest areas, instead of dividing the space up into squares, she looked at the position of each individual in space and time. Conclusion: this empirical approach produces better results than existing methodologies in terms of the consistency of the results, the robustness to the sampling frequency and the noise created by the simulations. “We should let the data do the talking,” claims the researcher.

Consider the context of the journey

Based on what she found, Nikolic raised questions about the methodological relationships typically used to characterize pedestrians’ movements. For example, walking speed is generally considered to be inversely proportional to density: the more people there are, the slower pedestrians walk. But the data collected suggest that this relationship is not deterministic. Some travelers weave through a crowd very quickly, while others stroll along even when there is nobody else around. This shows that walking speed does not depend solely on the density of people per square meter. Many other factors come into play, such as age, gender, time, the reason for traveling, whether the passenger has bags or not, and whether the individual is traveling alone or with others.

To take this range of factors into consideration, the researcher opted for a probability-based approach that also took account of the subject and context of the journey. She combined the information on individual trajectories collected from the sensors with the train timetables and infrastructure data. For the speed-density relationship, two types of pedestrians were identified: those who are more sensitive to congestion – and will therefore reduce their speed in a busy crowd – and those who are less sensitive. The model shows that the majority of passengers cannot in fact move at their ideal – or preferred – speed. “Unlike conventional models, ours produces results that more accurately reflect what we observe in real life,” says the researcher.

The model has been validated by empirical tests. Researchers have, for example, used it to predict whether users would welcome a change in the timetable. “This thesis shows how the ever-increasing amount of data can potentially be used to characterize and model pedestrian movements,” says Nikolic. “What makes the methodology even more interesting is that it can be used with other data sources and applied to other types of infrastructure such as museums, shopping centers, stadiums and streets.”

Data-driven fundamental models for pedestrian movement, Nikolić, Marija, Director: Bierlaire, Michel, 2017.