COVID-19 forecasting using shifted Gaussian Mixture Model with similarity-based estimation.

Area of Science:

• Epidemiology and Biostatistics
• Computational Modeling
• Data Science

Background:

• The COVID-19 pandemic significantly impacted global societies and economies.
• Accurate forecasting is crucial for disease control and mitigating severe cases.

Purpose of the Study:

• To develop a novel forecasting model for daily new COVID-19 cases.
• To improve prediction accuracy by leveraging international trend similarities.

Main Methods:

• Developed the Shifted Gaussian Mixture Model with Similarity-based Estimation (SGSE).
• Utilized Gaussian Mixture Model (GMM) for feature extraction from daily new case data.
• Identified countries with similar historical trends using time offsets and feature vectors.
• Introduced a trend similarity score for evaluating forecast accuracy.

Main Results:

• The SGSE model demonstrated superior forecasting accuracy compared to the ARIMA model (median trend similarity score 0.903-0.947 vs. 0.642).
• Evaluated in seven European countries, SGSE outperformed all other models in The European COVID-19 Forecast Hub.
• Accurate prediction of trends and plateaus was observed for multiple countries.

Conclusions:

• The SGSE model offers a highly accurate and reliable method for COVID-19 case forecasting.
• Leveraging cross-country trend similarities enhances predictive power.
• The developed trend similarity score provides a robust measure of forecast performance.