The feasibility of using machine learning to predict COVID-19 cases.

Area of Science:

• Epidemiology
• Public Health
• Data Science

Background:

• The COVID-19 pandemic exposed challenges in accurate epidemic data reporting, particularly in regions with limited healthcare infrastructure and testing capabilities.
• Underreporting of cases remains a critical issue, impacting global health crisis management and future preparedness.
• Machine learning offers a novel approach to identify discrepancies in reported versus predicted epidemic data.

Purpose of the Study:

• To evaluate the reliability of global COVID-19 incidence data, focusing on underdeveloped regions.
• To identify and quantify discrepancies between reported and machine learning-predicted COVID-19 cases.
• To underscore the importance of data accuracy for effective pandemic response.

Main Methods:

• Collected demographic, healthcare, economic, and testing data from March 2020 to September 2022.
• Employed diverse machine learning models including neural networks, decision trees, random forests, cross-validation, support vector machines, and logistic regression to predict COVID-19 incidence.
• Assessed model performance using testing accuracy metrics.

Main Results:

• Machine learning models achieved testing accuracy ranging from 55.92% to 65.50%.
• Neural network analysis identified significant underreporting of COVID-19 cases in many African countries compared to predicted rates.
• Discrepancies correlate with regions possessing limited COVID-19 testing capabilities.

Conclusions:

• There is a critical need for enhanced data accuracy and reporting, especially in resource-limited settings.
• Machine learning can effectively identify underreported epidemic cases, improving forecasting and response strategies.
• International collaboration and investment in testing infrastructure are vital for global health security.