COVINet: a deep learning-based and interpretable prediction model for the county-wise trajectories of COVID-19 in the United States

Journal of Applied Statistics, Journal Year: 2024, Volume and Issue: 52(5), P. 1063 - 1080

Published: Oct. 8, 2024

The devastating impact of COVID-19 on the United States has been profound since its onset in January 2020. Predicting trajectory epidemics accurately and devising strategies to curb their progression are currently formidable challenges. In response this crisis, we propose COVINet, which combines architecture Long Short-Term Memory Gated Recurrent Unit, incorporating actionable covariates offer high-accuracy prediction explainable response. First, train COVINet models for confirmed cases total deaths with five input features, compare Mean Absolute Errors (MAEs) Relative (MREs) against ten competing from CDC last four weeks before April 26, 2021. results show outperforms all MAEs MREs when predicting deaths. Then, focus most severe county each top 10 hot-spot states using COVINet. small predictions made 7 or 30 days March 23, 2023. Beyond predictive accuracy, offers high interpretability, enhancing understanding pandemic dynamics. This dual capability positions as a powerful tool informing effective prevention governmental decision-making.

Language: Английский

PAN-cODE: COVID-19 forecasting using conditional latent ODEs

Journal of the American Medical Informatics Association, Journal Year: 2022, Volume and Issue: 29(12), P. 2089 - 2095

Published: Sept. 1, 2022

The coronavirus disease 2019 (COVID-19) pandemic has caused millions of deaths around the world and revealed need for data-driven models spread. Accurate caseload forecasting allows informed policy decisions on adoption non-pharmaceutical interventions (NPIs) to reduce transmission. Using COVID-19 as an example, we present Pandemic conditional Ordinary Differential Equation (PAN-cODE), a deep learning method forecast daily increases in infections deaths. By using latent variable model, PAN-cODE can generate alternative trajectories based alternate adoptions NPIs, allowing stakeholders make manner. also estimation regions that are unseen during model training. We demonstrate that, despite less detailed data having fully automated training, PAN-cODE's performance is comparable state-of-the-art methods 4-week-ahead 6-week-ahead forecasting. Finally, highlight ability realistic outcome select US regions.

Language: Английский