COVID-19 Time Series Forecasting - Twenty Days Ahead

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• COVID-19 remains a significant global health concern, necessitating accurate forecasting for effective public health interventions and governmental decision-making.
• Predictive modeling plays a crucial role in anticipating disease spread and mortality, informing policy responses to the pandemic.

Purpose of the Study:

• To develop and evaluate an Artificial Neural Network (ANN) model for forecasting COVID-19 new cases and deaths.
• To assess the impact of incorporating mandatory mask usage as an input variable on the accuracy of the ANN prediction model.

Main Methods:

• Utilized Artificial Neural Networks (ANN) for time-series forecasting of COVID-19 epidemiological data.
• Compared prediction accuracy between models with and without the inclusion of mandatory mask usage data.
• Evaluated model performance for short-term forecasting (next twenty days) using data from Brazil and Portugal.

Main Results:

• The ANN model incorporating mandatory mask usage demonstrated improved accuracy in predicting COVID-19 cases and deaths within a twenty-day forecast horizon.
• For Brazil, the model achieved a forecasting error of 24.7% for cumulative infection cases and 1.6% for deaths.
• For Portugal, the forecasting error was 37.9% for cumulative infection cases and 33.8% for deaths.

Conclusions:

• The integration of public health interventions, such as mandatory mask usage, into epidemiological models can enhance forecasting accuracy.
• ANN models show promise for short-term prediction of COVID-19 trends, aiding in public health strategy development.
• The study highlights the importance of considering mitigation measures in predictive models for infectious diseases.