Meta-Learning for County-Level COVID-19 Prevalence Prediction from Wastewater Surveillance Data

Abstract

This presentation outlines a research proposal for a meta-learning framework designed to predict infectious disease outbreaks across cities using wastewater-based epidemiology (WBE) data. The task is to train a model that captures generalizable outbreak dynamics from multiple cities and can be efficiently adapted to new locations with minimal data. The initial scope focuses on COVID-19, leveraging well-established and continuously updated datasets such as the CDC National Wastewater Surveillance System, EU Sewage Sentinel, and NORMAN-SCORE. Meta-learning is motivated as a solution to persistent challenges in WBE, including site-specific measurement differences, limited data availability, and poor cross-city generalization. The proposed methodology includes multi-city training, city-level adaptation, and comparison against city-specific baseline models to assess forecasting accuracy. The presentation highlights both feasibility and realism concerns, emphasizing that while data incompleteness remains a limiting factor, meta-learning provides a principled approach to improving robustness and transferability in wastewater-driven outbreak prediction.