Auditing the fairness of the US COVID-19 forecast hub’s case prediction models
Published in PLOS One, 2025
Abstract
Systematic audit of fairness and accuracy across 36 COVID-19 forecasting models from the CDC Forecast Hub, revealing persistent disparities in forecast errors across demographic groups.
Key Findings
- Forecast errors systematically higher in counties with larger Hispanic populations and rural communities
- Deep-learning and ensemble models showed more balanced error profiles than conventional approaches
- Need for standardized race/ethnicity fields and improved rural sampling in public health surveillance
- Pattern persists across model families: compartmental, statistical, ensemble, and deep-learning approaches
Recommended Citation
Abrar, S. M., Awasthi, N., Smolyak, D., Sigalo, N., & Martinez, V. F. (2025). Auditing the fairness of the US COVID-19 forecast hub’s case prediction models. PloS one, 20(4), e0319383.
Recommended citation: Abrar, S. M., Awasthi, N., Smolyak, D., Sigalo, N., & Martinez, V. F. (2025). Auditing the fairness of the US COVID-19 forecast hub case prediction models. PloS one, 20(4), e0319383.
Download Paper