Forecasting COVID-19 outbreak progression using hybrid polynomial-Bayesian ridge regression model.

Area of Science:

• Epidemiology
• Machine Learning
• Computational Statistics

Background:

• The COVID-19 pandemic, caused by SARS-CoV-2, presented unprecedented global health challenges.
• Accurate forecasting of infected cases is crucial for effective public health response and resource allocation.
• Existing forecasting methods often struggle with uncertainty and incorporating new data efficiently.

Purpose of the Study:

• To propose a novel hybrid machine learning model for predicting COVID-19 cases.
• To develop a model that accurately forecasts infected cases while quantifying prediction uncertainty.
• To create a flexible mathematical model capable of incorporating prior knowledge and new data.

Main Methods:

• A hybrid model combining Bayesian Ridge Regression with an n-degree Polynomial was developed.
• Probabilistic distributions were used for estimating the dependent variable, moving beyond traditional deterministic methods.
• L² (Ridge) Regularization was implemented to prevent model overfitting and enhance generalization.
• The model incorporates prior knowledge and posterior distributions for efficient data assimilation.

Main Results:

• The hybrid model demonstrated high accuracy in predicting COVID-19 cases.
• The model successfully quantified the uncertainty associated with its predictions.
• Case studies in the United States, Italy, and Spain validated the model's forecasting capabilities.
• Forecasts were based on publicly available data up to May 11, 2020.

Conclusions:

• The proposed hybrid model offers a robust approach to COVID-19 forecasting.
• The model's ability to handle uncertainty and incorporate new data makes it valuable for public health.
• Further research and evolution of the model are recommended for ongoing pandemic management.