When Do We Need Massive Computations to Perform Detailed COVID-19 Simulations?

Area of Science:

• Epidemiology
• Computational Biology
• Machine Learning

Background:

• The COVID-19 pandemic necessitated rapid decision-making for intervention strategies.
• Computer simulations aid in predicting disease outcomes but demand substantial computational resources.
• Existing simulation methods present a time and processing power challenge for officials.

Purpose of the Study:

• To investigate the efficacy of machine learning meta-models in predicting COVID-19 disease trajectories.
• To determine if training on limited simulation data can yield accurate predictions.
• To assess the cost-effectiveness of machine learning models compared to traditional simulations.

Main Methods:

• Utilized four established agent-based models (ABMs) for COVID-19.
• Developed decision tree regression models for each ABM.
• Compared the predictive accuracy of machine learning meta-models against full ABM results.

Main Results:

• Accurate meta-models were achieved with minimal data (25%) for ABMs without strong interventions.
• Root-mean-square error (RMSE) was comparable between models trained on partial and full datasets.
• ABMs with strong interventions required significantly more training data (≥60%) for comparable accuracy.

Conclusions:

• Machine learning meta-models can effectively predict disease trajectories in certain scenarios.
• This approach can significantly reduce the computational cost and time required for epidemiological modeling.
• The feasibility of using ML meta-models is dependent on the complexity of simulated interventions.