SIRVD-DL: A COVID-19 deep learning prediction model based on time-dependent SIRVD

Area of Science:

• Epidemiology
• Infectious Disease Modeling
• Artificial Intelligence in Healthcare

Background:

• COVID-19 presents significant global health challenges requiring accurate transmission prediction.
• Existing artificial intelligence (AI) models for COVID-19 prediction often lack the ability to capture temporal transmission dynamics.
• Mathematical models like SIRVD are crucial for understanding infectious disease spread but can be enhanced with AI.

Purpose of the Study:

• To develop an improved COVID-19 prediction model by integrating deep learning with a time-dependent SIRVD framework.
• To enhance the accuracy and interpretability of infectious disease forecasting.
• To address the limitations of purely AI-based prediction methods in capturing time-varying transmission patterns.

Main Methods:

• Proposed a hybrid model fusing deep learning (LSTM) with the time-dependent SIRVD (Susceptible-Infectious-Recovered-Vaccinated-Deceased) compartmental model.
• Utilized deep learning to forecast key parameters within the SIRVD mathematical model.
• Analyzed COVID-19 data from January 15, 2021, to May 27, 2021, across nine countries during mass vaccination.

Main Results:

• The hybrid deep learning-SIRVD model demonstrated a 50% improvement in single-day COVID-19 predictions compared to pure deep learning approaches.
• The model proved effective for both short- and medium-term forecasting of infectious disease trends.
• The integrated approach enhanced the interpretability and robustness of the COVID-19 predictions.

Conclusions:

• Combining deep learning with mathematical models like SIRVD offers a more robust and accurate method for predicting infectious disease transmission.
• This hybrid approach overcomes limitations of standalone AI or compartmental models for dynamic disease forecasting.
• The developed model provides a valuable tool for understanding and managing the evolving COVID-19 pandemic.