Data driven estimation of novel COVID-19 transmission risks through hybrid soft-computing techniques.

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• The COVID-19 pandemic emerged as a global health crisis, spreading rapidly across 201 countries.
• As of April 2020, over 1.1 million infections and 59,000 deaths were recorded worldwide.
• Understanding pandemic spread dynamics and forecasting case numbers are critical for effective public health responses.

Purpose of the Study:

• To develop and evaluate soft-computing hybrid models for forecasting COVID-19 transmissibility.
• To predict the increase and maximum number of virus-infected cases in various regions.
• To provide data-driven insights for healthcare resource allocation and policy-making.

Main Methods:

• Utilized a hybrid approach combining wavelet decomposition with a neuronal-fuzzification technique.
• Developed a wavelet-based forecasting model for short-term predictions (5-10 days).
• Employed data-based prediction via moving average for longer-term forecasts (50-60 days).

Main Results:

• Wavelet-based models demonstrated higher accuracy for short-term COVID-19 case forecasting in China, India, and the USA.
• Longer-term predictions using moving averages showed lower accuracy compared to wavelet-based hybrids.
• Performance metrics included MASE (0.06-5.76) and sMAPE (0.15-1.97), indicating model effectiveness and outlier resistance.

Conclusions:

• Hybrid soft-computing models, particularly wavelet-based approaches, are effective for short-term COVID-19 forecasting.
• Accurate forecasting aids in optimizing healthcare facility allocation and serves as an early-warning system.
• Data-driven analysis provides crucial insights into viral transmission, helping to mitigate panic and stigma associated with the pandemic.