ALeRT-COVID: Attentive Lockdown-awaRe Transfer Learning for Predicting COVID-19 Pandemics in Different Countries

Area of Science:

• Epidemiology
• Machine Learning
• Public Health

Background:

• COVID-19 spread necessitates public health interventions like lockdowns, impacting economies.
• Accurate prediction of epidemic trajectories is vital for informed policy decisions regarding lockdown measures.

Purpose of the Study:

• To develop and evaluate ALeRT-COVID, a transfer learning model using attention-based RNNs for predicting COVID-19 trends.
• To incorporate lockdown measures as a predictor within the model to assess their impact on epidemic progression.

Main Methods:

• Utilized a transfer learning approach, training a source model on pre-defined countries and adapting it to target countries.
• Employed an attention-based recurrent neural network (RNN) architecture to weigh past confirmed cases' influence on future trends.
• Integrated lockdown status as a key predictor variable in the forecasting model.

Main Results:

• Transfer learning demonstrated significant benefits, particularly for countries in the early stages of the pandemic.
• The inclusion of lockdown predictors and the attention mechanism substantially improved ALeRT-COVID's prediction performance.
• One-week ahead predictions indicated the continued necessity of lockdown measures in several nations.

Conclusions:

• ALeRT-COVID offers an effective method for predicting COVID-19 epidemic trends, especially when incorporating lockdown data.
• The model's transfer learning capability enhances predictions for countries with limited initial data.
• Findings support the strategic use of lockdown measures, aiding policymakers in making informed decisions for pandemic management.