Forecasting Epidemics: From Ensembles to Graph Neural Networks

Epidemics are public health emergencies that demand rapid detection, real-time forecasting, and effective intervention strategies. Forecasting epidemic dynamics is especially challenging due to issues such as data quality, human behavioral variability, and evolving pathogen characteristics. Recent years have seen significant advances in computational modeling to better forecast disease spread and support timely public health responses.

In this talk, I will discuss the key challenges of real-time epidemic forecasting and our ongoing efforts to develop robust modeling frameworks. I will focus on data-driven approaches, including ensemble methods and graph neural network-based spatiotemporal models. I will highlight strategies for model training, integration of auxiliary data sources (e.g., internet search trends, wastewater signals), and evaluation techniques grounded in statistical and information-theoretic principles. Finally, I will introduce tools developed for public health analysts and describe initiatives for benchmarking epidemic forecasting models.

Although the focus is on epidemic forecasting, many of the techniques discussed are domain-agnostic and applicable to a wide range of machine learning tasks.