Deep learning forecasting using time-varying parameters of the SIRD model for Covid-19

Area of Science:

• Epidemiology
• Computational Biology
• Data Science

Background:

• Accurate epidemiological models are crucial for managing the novel coronavirus pandemic.
• The basic reproduction number is a key metric for understanding disease spread dynamics.
• Existing models may lack real-time forecasting capabilities and time-dependent parameter estimation.

Purpose of the Study:

• To develop a novel algorithm for accurate, real-time epidemiological forecasting of the novel coronavirus pandemic.
• To integrate the susceptible-infected-recovered-dead (SIRD) model with long short-term memory (LSTM) neural networks.
• To enable time-dependent estimation of epidemiological parameters like contact and deceased rates.

Main Methods:

• Developed a hybrid algorithm combining the SIRD epidemiological model with LSTM neural networks.
• Incorporated real-time mobility data and positive test rates into the prediction model.
• Utilized a vaccination model to account for public health interventions.

Main Results:

• The novel algorithm demonstrates improved forecasting accuracy for epidemiological parameters.
• The model allows for real-time, time-dependent estimation of contact and deceased rates.
• Integration of mobility and vaccination data enhances the model's predictive power.

Conclusions:

• The proposed LSTM-SIRD algorithm offers a significant advancement in real-time pandemic forecasting.
• Accurate prediction of epidemiological parameters aids in informed decision-making for public health.
• Leveraging mobility and vaccination data is essential for capturing behavioral dynamics and intervention impacts.