Analysis of mobility based COVID-19 epidemic model using Federated Multitask Learning.

Area of Science:

• Epidemiology
• Machine Learning
• Data Science

Background:

• Federated Learning (FL) faces challenges in distributing global models due to heterogeneous local data distributions.
• Personalized models are crucial for accurate disease prediction in distributed environments.

Purpose of the Study:

• To propose a personalized model using federated multitask learning (FMTL) for predicting COVID-19 infection rates in the USA.
• To analyze hospital bed availability using a mobility-based SEIR model with an added constraint.

Main Methods:

• Utilized real-time mobility data from five US states (2020-2021).
• Employed a federated learning environment with each state as a node, implementing a linear regression model at each node.
• Developed a mobility-based SEIR model for infection rate prediction and bed availability analysis.

Main Results:

• Observed correlations in COVID-19 case numbers across states despite differing spread rates.
• Colorado state showed the lowest root-mean-square percentage error in COVID-19 case prediction, while Minnesota showed the highest.
• The mobility-based SEIR model simulation indicated approximately 400 days to reach extinction without vaccination or social distancing.

Conclusions:

• Federated multitask learning effectively creates personalized models for predicting infectious disease dynamics.
• Mobility data integrated with SEIR models provides valuable insights into disease spread and healthcare resource management.
• Public health interventions like vaccination and social distancing are critical for mitigating pandemic duration.