Machine learning mathematical models for incidence estimation during pandemics

Area of Science:

• Epidemiology
• Computational Biology
• Machine Learning

Background:

• Accurate infectious disease incidence data is crucial for epidemic control.
• Under-reporting is common due to limited testing, especially with asymptomatic cases.
• Real-time incidence estimation is vital for timely public health interventions.

Purpose of the Study:

• To develop a machine learning approach for real-time pandemic incidence estimation.
• To identify parsimonious, closed-form mathematical models for incidence prediction.
• To validate the model's accuracy using COVID-19 data from multiple countries.

Main Methods:

• Utilized Bayesian symbolic regression to automatically derive mathematical models.
• Incorporated reported case counts and overall test rates as input features.
• Validated models using daily COVID-19 incidence data from nine countries.

Main Results:

• The machine learning models accurately predicted daily infectious disease incidence.
• A single, unified model demonstrated superior parsimony and predictive power across diverse countries compared to country-specific models.
• The approach effectively addresses under-reporting by integrating testing rates.

Conclusions:

• Machine learning, specifically Bayesian symbolic regression, offers a powerful tool for real-time incidence modeling.
• A universal model can effectively capture pandemic dynamics across different regions.
• This method provides a valuable, accurate tool for public health decision-making during epidemics.