Dynamic Hybrid Model to Forecast the Spread of COVID-19 Using LSTM and Behavioral Models Under Uncertainty

Area of Science:

• Epidemiology
• Computational modeling
• Artificial intelligence

Background:

• Accurate prediction of coronavirus disease 2019 (COVID-19) regional spread is crucial for effective public health interventions.
• Uncertainty from confounding variables significantly impacts COVID-19 transmission dynamics.

Purpose of the Study:

• To develop and validate a novel hybrid model for accurate regional COVID-19 spread prediction.
• To enhance prediction accuracy by incorporating dynamic behavioral models alongside artificial intelligence.

Main Methods:

• A hybrid model integrating Long Short-Term Memory (LSTM) networks with dynamic behavioral models was developed.
• The model was optimized using a genetic algorithm, considering regional properties for each of the top ten most-affected countries.
• The model accounts for multiple factors influencing virus spread and uncertainty.

Main Results:

• The hybrid model demonstrated high accuracy in replicating test data for COVID-19 cases and deaths.
• It successfully predicted the daily behavior of the pandemic under uncertainty.
• The hybrid model showed superior performance compared to a standalone LSTM model, especially with data limitations.

Conclusions:

• The proposed hybrid model offers accurate short-to-medium term predictions of COVID-19 daily spread.
• This predictive capability supports effective policy assessment, planning, and decision-making for pandemic management.
• The model's ability to handle uncertainty and regional variations enhances its practical utility.