Improving prediction of COVID-19 evolution by fusing epidemiological and mobility data

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• The COVID-19 pandemic's consequences extend beyond healthcare, necessitating effective control measures.
• Predicting pandemic evolution is challenging due to health system strain and unpredictable human behavior.
• Current policy decisions rely on infection trends, which are difficult to forecast accurately.

Purpose of the Study:

• To evaluate diverse models for early COVID-19 pandemic evolution prediction.
• To develop a decision support system for policy-makers.
• To improve the accuracy of short-to-medium term pandemic forecasting.

Main Methods:

• Utilized artificial neural networks (LSTM, GRU) and statistical models (AR, ARIMA).
• Employed consensus strategies to ensemble multiple models for enhanced performance.
• Developed a multivariate model incorporating Google mobility data for trend change prediction.

Main Results:

• Achieved high accuracy in predicting 14-day case increases (CI) in a Spanish case study.
• Demonstrated strong performance in scenarios with and without trend changes.
• Reported metrics include R-squared of 0.93, RMSE of 4.16, and MAE of 1.08.

Conclusions:

• The proposed ensemble and multivariate models provide accurate early predictions of COVID-19 evolution.
• Integrating mobility data significantly enhances forecasting of trend changes.
• The developed system can aid policy-makers in managing the pandemic and its socioeconomic impacts.