Interaction-Temporal GCN: A Hybrid Deep Framework For Covid-19 Pandemic Analysis

Area of Science:

• Computational epidemiology
• Data science
• Graph neural networks

Background:

• The Covid-19 pandemic poses a significant global health threat, necessitating advanced analytical tools.
• Existing Covid-19 data, often with spatial-temporal properties, can be leveraged for predictive modeling.
• Understanding the complex spatial-temporal dynamics of disease spread is crucial for effective containment.

Purpose of the Study:

• To propose a novel framework, the Interaction-Temporal Graph Convolution Network (IT-GCN), for analyzing pandemic data.
• To model the spatial-temporal linkages and underlying interaction topology within Covid-19 transmission data.
• To achieve accurate short-term prediction of daily Covid-19 infected cases.

Main Methods:

• Developed the Interaction-Temporal Graph Convolution Network (IT-GCN) framework.
• Integrated Autoregressive Integrated Moving Average (ARIMA) with Graph Convolutional Networks (GCN).
• Constructed graph nodes using ARIMA parameterization to represent inter-city pandemic interactions.
• Utilized Covid-19 daily case data from the United States for experimental validation.

Main Results:

• The IT-GCN framework successfully captured comprehensive interaction-temporal topology in pandemic data.
• IT-GCN achieved well-performed short-term prediction of daily Covid-19 infected cases in the US.
• The proposed method outperformed state-of-the-art baselines in terms of Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE).

Conclusions:

• The IT-GCN is a valid and effective method for forecasting Covid-19 daily infected cases.
• The framework's ability to predict disease spread can significantly assist in improving public health containment policies.
• IT-GCN shows potential for application in analyzing other related complex time-series data.