Forecasting COVID-19 pandemic using optimal singular spectrum analysis.

Area of Science:

• Epidemiology
• Time Series Analysis
• Computational Statistics

Background:

• The COVID-19 pandemic necessitates accurate forecasting models for public health.
• Existing time series models have varying performance in predicting pandemic trends.

Purpose of the Study:

• To evaluate Singular Spectrum Analysis (SSA) for forecasting COVID-19 daily confirmed cases, deaths, and recoveries.
• To compare SSA's forecasting accuracy against established time series methods.
• To propose an algorithm for optimizing SSA parameters.

Main Methods:

• Developed an algorithm for optimal SSA parameter selection (window length, leading components).
• Compared SSA (vector and recurrent forecasting) with ARIMA, ARFIMA, Exponential Smoothing, TBATS, and NNAR.
• Utilized Root Mean Squared Error (RMSE) for model accuracy assessment.
• Applied the best model to forecast 40 days ahead using Johns Hopkins University CSSE data.

Main Results:

• No single forecasting model consistently outperformed others across all countries and timeframes.
• SSA demonstrated viability by frequently outperforming competing models.
• Forecasts were generated for the top ten most affected countries until October 29, 2020.

Conclusions:

• SSA is a viable technique for forecasting COVID-19 epidemiological data.
• Comparative analysis highlights the strengths and weaknesses of various forecasting methods.
• Accurate forecasting aids in predicting disease behavior and informing public health decisions.