Dynamic SEIR-GNN Using Individualized Data For Risk Prediction During The Pandemic

Abstract

Predicting the impact of a pandemic on a population is a question that has been debated for decades and remains unresolved. In our proposed study, we adopted a strategy that leverages the mechanistic understanding of compartmental models and the high adaptability of deep learning models to design a more effective mathematical model for pandemic prediction. This mathematical model is designed to predict the severity of future pandemic outbreaks based on geographic location and the type of hotspots individuals access daily, incorporating both human transmission and environmental transmission variables to enhance prediction accuracy. The proposed model uses a SEIR compartmental model with a multilayered GNN algorithm to predict the model parameters. The proposed multilayered GNN comprises individual agents, clusters based on the type of importance, and significant geographical clusters, each representing individual nodes and layers within the GNN. The edge weight between nodes is updated based on risk scores, which are calculated based on the likelihood of an agent carrying a pathogen while interacting with that region. Using this model, it is possible to predict the outcome of a pandemic by changing various parameters and conditions. This adaptability will be invaluable when planning for long-term pandemic predictions, particularly when the initial parameters of transmission rates and recovery rates for individuals remain unclear. The proposed model yields better prediction accuracy, with a Root Mean Square Error (RMSE) value of 2.85 and a Mean Absolute Error (MAE) of 2.49, outperforming standalone compartmental models. Policymakers and governmental agencies can use this hybrid model to frame various control measures during the pandemic.