Projects Research collaboration 2022 to 2024
Early Warning System for Infectious Diseases
Spatiotemporal modelling and NLP over news and social posts to detect outbreak signals early, as part of the Mathematics for Public Health initiative.
The problem
Official surveillance data is reliable and slow. Text on the open web is fast and extremely noisy. An early-warning system has to extract a usable signal from the second without inheriting all of its noise, and then place that signal in space and time well enough for a public-health body to act on it.
Approach
The pipeline combines document-level classification of disease mentions with spatiotemporal models that pool information across neighbouring regions and recent weeks. The statistical question underneath is a hard one: how do you calibrate an alarm threshold when your data-generating process includes media attention as a confounder?
Placeholder description. Edit
_projects/early-warning-system.mdto add results, figures or links once they are public.