From the European Championship to the Pandemic: How Mobile Data Provides Insights into Population Contact Patterns
The Data Science Team at NET CHECK is exploring what insights can be gained from mobile data as part of various research projects. With a total of four topics, the team, led by Machine Learning Team Leader Steven Schulz, has been invited to the prestigious NetMob Conference in Washington.
What is the NetMob Conference about?
S.Schulz: The NetMob Conference brings together industry and academic participants from around the world who work with mobile data. In short presentations and posters, current research findings are presented to promote collaboration and networking for new projects.
Since 2010, the conference has been held every two years at different locations and has become an important meeting point for all topics and players related to mobile data. This year, it will be held at the World Bank Headquarters in Washington, D.C. We will present the results of two research projects, DAKI-FWS* and OptimAgent**, as well as an internal project in the telecommunications sector.
What is the data basis for your research?
S.Schulz: For our analyses, we use NET CHECK’s crowdsourcing data, which is mobile data based on a panel of approximately 1.1 million daily users. The data contains anonymised location and time information from mobile phones, from which we can, among other things, infer how people move around the city or how often phones—and therefore people—come into contact. Based on this, we have investigated interesting questions in the areas of mobility and contact behavior.
What questions did you explore in the area of mobility?
S.Schulz: Among other things, we analysed the movement patterns of the population in Berlin and identified what communities form. We showed how historical developments in the city of Berlin, particularly the division between East and West, still unconsciously affect people’s daily lives today.
How does the former division of Berlin still affect the city today, over 35 years after the fall of the Wall?
S.Schulz: We clearly see a continued separation of East and West communities, stemming from the historical division of the city. One consequence of this division is that there are still fewer transportation links between the eastern and western parts of the city. However, these few connections are used all the more intensively.
These findings can be very useful for infrastructure planning, as they indicate a need for more connections between the two sides of the city, which could be addressed, for example, by additional bus and train routes.
What questions did you explore regarding contact behavior?
S.Schulz: On one hand, we developed a prediction tool for infection numbers that combines a contact index from mobile data with Bayesian mixed-media models, originally developed by Google to predict sales numbers. The contact index shows how many contact points there are within the population and reflects the contact behavior relevant to epidemics.
We remember the reproduction number (R-value) from the COVID-19 pandemic, which estimates how many people, on average, are infected by a single person. The contact index enables us to predict the development of the R-value and, thus, the change in infection numbers over a specific period.
We also present a study in which we analyzed the contact behavior of football fans during the 2024 European Football Championship in Germany.
What exactly did you investigate regarding the European Championship?
S.Schulz: On a national, city, and stadium level, we determined how many contact points there were, focusing particularly on points of interest such as bars, restaurants, and fan zones near the football matches.
We observed that contact density increased during bad weather. Furthermore, contact density was higher during matches that were particularly significant or had a large fan base. We saw this not only in the cities where the games were played but across Germany. So, during bad weather and exciting matches, people across the country came together more closely. The results are very interesting in an epidemiological context because they indicate an increased risk of infection in such situations.
“The economic value and societal relevance of mobile data, as generated by NET CHECK, go far beyond the telecommunications sector.”
(Steven Schulz, Head of Machine Learning at NET CHECK)
What is the relevance of your research findings?
S.Schulz: The projects show that both the economic utility and the societal relevance of mobile data, as generated by NET CHECK, extend far beyond the telecommunications sector. The use cases address challenges in public health, infrastructure, urban planning, and disaster prevention.
What advantage does mobile data have over other methods?
S.Schulz: The key lies in obtaining an accurate, statistical picture of the population over long periods and in near-real-time, while preserving individual anonymity. Through the crowdsourcing approach, we were able to represent 1% of the German population, which provided very representative results.
Compared to traditional methods such as surveys (e.g., regarding contacts during the pandemic) or headcounts (e.g., to determine the occupancy of public transport), the advantage is the ability to quickly reach large groups of people through popular apps without requiring active data input. This approach will be very helpful in the future for predicting infection numbers and preventing the rapid spread of new virus variants.
Which results personally surprised you the most?
S.Schulz: We have shown that the emergence of new virus variants can be predicted without biological methods, using only GPS data. This means that in future pandemics, it could be possible to respond more quickly to such changes. Additionally, the crowdsourcing approach allows us to obtain a highly accurate and representative sample of the population, all while maintaining anonymity and without the need for manual data entry. Factors like human error due to memory lapses or incorrect entries are also eliminated.
*DAKI-FWS stands for “Data and AI-Based Early Warning System for Stabilizing the German Economy.” More information: https://daki-fws.de/
**The primary goal of OptimAgent is to develop an agent-based model specifically tailored to the German population, which will serve as decision support during pandemics. This model is being developed by scientists from 13 different institutions and 6 sub-projects. More information: https://webszh.uk-halle.de/optimagent/
