DeepEVD: Integrating Epidemiological data into deep learning frameworks based on spatio-temporal feature learning for...

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Traditional epidemiological models often fail to incorporate human mobility, a key factor in disease transmission.
• Ebola Virus Disease (EVD) outbreaks are significantly influenced by population movement patterns.
• Accurate forecasting of EVD outbreaks is critical for effective public health interventions.

Purpose of the Study:

• To develop and validate DeepEVD, a novel framework for forecasting EVD outbreaks by integrating human mobility data.
• To enhance the accuracy of EVD outbreak prediction compared to existing epidemiological models.
• To provide actionable insights for EVD prevention and control strategies.

Main Methods:

• DeepEVD framework utilizes diverse human mobility data (phone records, GPS, social media).
• Employs Graph Convolutional Networks (GCN) and Long Short Term Memory (LSTM) for spatio-temporal feature extraction.
• Validated on real-world EVD outbreak datasets from West Africa (2014-2016) and Sierra Leone (2015-2016).

Main Results:

• DeepEVD demonstrated a 5%-10% reduction in forecasting errors compared to baseline methods.
• Ablation studies confirmed the impact of different data sources and feature extraction techniques on predictive accuracy.
• The framework achieved state-of-the-art performance in EVD outbreak forecasting.

Conclusions:

• DeepEVD offers a significant advancement in predicting EVD outbreaks by incorporating human mobility.
• The framework provides valuable insights for enhancing EVD surveillance and response efforts.
• The study highlights the potential of integrating big data and machine learning in infectious disease epidemiology.