As machine learning continues to revolutionize various scientific domains, its impact on epidemic time series forecasting has become increasingly significant. This talk examines how advanced machine learning methods can address several pressing challenges in epidemic forecasting, including capturing spatiotemporal disease dynamics, coping with limited data, and developing scalable tools for research and deployment.
I will first present a graph neural ODE framework for modeling the continuous spread of infectious diseases across regions. I will then show how pretraining on large-scale epidemic data can improve forecasting accuracy and enhance generalization across heterogeneous outbreak settings. Finally, I will introduce EpiLearn, our modular open-source Python toolkit for machine learning in epidemic modeling, which supports forecasting and source detection through unified pipelines for datasets, transformations, simulation, benchmarking, and visualization. I will conclude by highlighting several promising directions for future research in machine learning for epidemic forecasting.