Can AI predict the next disease outbreak before authorities become aware?
Scientists at IT:U have developed an early warning system that can spot disease outbreaks through geospatial information early on.
The scientists from the Geosocial Artificial Intelligence team analyze real-time public data from social media, such as from people posting about their symptoms when they feel sick or are stuck in bed.
Based on this information, they can zoom into areas of concern, oftentimes before official data becomes available.
“Traditional epidemiological data carries a built-in time delay. By the time a patient feels sick, visits a doctor, has a test processed, and the data is officially collected and published, 10 to 14 days have usually passed.”
Bernd Resch | IT:U, Professor of Geosocial Artificial Intelligence.
Because people post on social media almost instantly when they start feeling ill, he says “our system catches the upward curve of an outbreak 10 to 14 days ahead of official health statistics. It essentially acts as a functional early warning system”.
As the system spots health threats early on, it would allow authorities to take immediate action.
Working with Harvard University in the United States, the system’s accuracy has been investigated on different continents and for different diseases, including Covid-19 in Austria and the USA, dengue fever outbreaks in Brazil, and malaria in the DRC.
“By analyzing these posts, we create a spatial-temporal view (where and when) of a disease spreading. When we correlate this user data with official health data (like laboratory case counts or hospitalizations), we find that our social media ‘nowcasting’ is reliable, particularly when coupled with other data sources.”
Data density is key
Professor Resch has been invited to give a keynote lecture at the Association of Geographic Information Laboratories in Europe, or Agile 2026, conference in Estonia this week to talk about generating actionable insights from geospatial data.
The conference is one of the leading geoinformatics conferences worldwide.
He stresses data density is important for his research because, “we need a high enough volume of georeferenced data over time to generate a statistically robust signal”.
“When a high volume of people in a specific geographic location post about a singular topic, our AI algorithms evaluate the statistical confidence of that data. If the confidence is high, the AI identifies it as a verified real-world event, allowing us to detect anomalies in real time.”
Bernd Resch, IT:U Professor of Geosocial Artificial Intelligence.
Next to trying to spot the first wave of disease outbreaks, the team is also working with colleagues in Germany, Brazil and the USA on identifying mosquito breeding sites to help prevent outbreaks of illnesses, such as dengue fever or malaria.
They have also used and tested this methodology in other areas, such as flood response, human migration, and urban planning.