An Epidemiological Neural Network Exploiting Dynamic Graph Structured Data Applied to the COVID-19 Outbreak.

Area of Science:

• Epidemiology
• Machine Learning
• Network Science

Background:

• The COVID-19 pandemic highlighted the need for accurate epidemic prediction models.
• Existing models often struggle to incorporate real-world mobility and place-based data.
• Dynamic factors influencing disease spread require advanced analytical approaches.

Purpose of the Study:

• To propose a novel machine learning framework for estimating epidemiological model parameters.
• To integrate static and dynamic features of places into epidemic modeling.
• To enhance the prediction accuracy of epidemiological models like SIR and SIRD.

Main Methods:

• A machine learning framework combining Graph Convolutional Neural Networks (GCNs) and Long Short-Term Memory (LSTM) networks.
• Modeling mobility data as a graph series to capture spatial and temporal dynamics.
• Estimating key epidemiological parameters such as contact and recovery rates.

Main Results:

• The proposed framework successfully inferred parameters for SIR and SIRD models.
• Evaluation using COVID-19 data from Italy demonstrated the model's predictive capabilities.
• The approach effectively utilizes place-based static and dynamic features for parameter estimation.

Conclusions:

• The novel machine learning framework offers a powerful tool for epidemiological parameter estimation.
• Integrating mobility data via GCNs and LSTMs improves the accuracy of epidemic spread predictions.
• This methodology can be applied to various epidemiological models and infectious disease outbreaks.