Spatio-temporal epidemic forecasting using mobility data with LSTM networks and attention mechanism

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Infectious disease outbreaks significantly disrupt global socio-economic stability.
• Accurate short-term forecasting is vital for effective public health policy and healthcare management.
• COVID-19 highlighted the need for advanced predictive tools to manage pandemics.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for short-term infectious disease forecasting.
• To assess the model's performance using COVID-19 data from Japan.
• To explore the impact of mobility data on forecast accuracy.

Main Methods:

• A deep learning model integrating Long Short-Term Memory (LSTM) networks and a multi-head attention mechanism.
• Training on publicly available COVID-19 confirmed cases and hospitalization data.
• Weekly forecasting of trends over 1-4 week horizons, with performance evaluation using multiple error metrics.

Main Results:

• The proposed model significantly outperforms baseline methods in short-term forecasting accuracy.
• Lower error rates were observed across multiple metrics for weekly predictions.
• Incorporating mobility data enhanced predictive accuracy, particularly for longer-term forecasts.

Conclusions:

• The novel deep learning model offers a powerful tool for short-term infectious disease trend prediction.
• The model's ability to integrate mobility data improves its capacity to capture spatio-temporal dynamics.
• This approach can aid decision-making, disease control strategies, and pandemic impact mitigation.