Reservoir computing on epidemic spreading: A case study on COVID-19 cases.

Area of Science:

• Computational epidemiology
• Complex systems science
• Machine learning applications

Background:

• Predicting infectious disease spread is crucial for public health interventions.
• Traditional epidemiological models often require detailed parameterization.
• Data-driven approaches offer complementary predictive capabilities.

Purpose of the Study:

• To develop and evaluate an echo state network (ESN) for predicting disease transmission dynamics.
• To assess the ESN's ability to forecast infection trends using historical and cross-location data.
• To explore the ESN's performance without requiring explicit epidemiological rate parameters.

Main Methods:

• Utilized a reservoir computing approach with an echo state network (ESN).
• Trained and tested the ESN on synthetic data from susceptible-infected-recovery (SIR) models.
• Validated the ESN's predictive performance using real-world COVID-19 pandemic data from various countries.
• Employed time-evolving datasets from numerous locations to capture disease progress history.

Main Results:

• The ESN successfully predicted infection trends in synthetic data for 4-5 weeks by leveraging data from uncorrelated patches.
• For real-world COVID-19 data, the ESN predicted trends for up to 3 weeks in targeted locations.
• The model's accuracy depended on historical disease progression data rather than detailed epidemiological parameters.
• A modified ESN scheme was developed for future disease spread forecasting.

Conclusions:

• Echo state networks provide a robust framework for predicting infectious disease spread using time-series data.
• The ESN's data-driven nature reduces the need for complex epidemiological parameter estimation.
• This approach shows promise for short-term epidemiological forecasting and can be extended for longer-term predictions.