Forecasting West Nile Virus With Graph Neural Networks: Harnessing Spatial Dependence in Irregularly Sampled...

Area of Science:

• Environmental science
• Epidemiology
• Computer science

Background:

• Machine learning is increasingly used for geospatial environmental issues.
• Existing methods for mosquito-borne disease forecasting often neglect spatial data structures.
• Accurate forecasting is crucial for mosquito surveillance and disease control.

Purpose of the Study:

• To develop and evaluate a spatially aware graph neural network for West Nile virus presence forecasting.
• To improve mosquito surveillance and abatement strategies in Illinois.
• To demonstrate the efficacy of graph neural networks for irregularly sampled geospatial data.

Main Methods:

• Application of a spatially aware graph neural network model utilizing GraphSAGE layers.
• Forecasting the presence of West Nile virus in Illinois.
• Comparison against baseline methods like logistic regression, XGBoost, and fully-connected neural networks.

Main Results:

• The graph neural network model demonstrated superior performance in forecasting West Nile virus presence.
• The spatially aware approach effectively incorporates underlying spatial data structures.
• Graph neural networks outperformed traditional machine learning models on this geospatial task.

Conclusions:

• Spatially aware graph neural networks offer a powerful tool for environmental and disease forecasting.
• This approach can significantly enhance mosquito surveillance and West Nile virus abatement efforts.
• Graph neural networks provide a more effective solution for analyzing complex, irregularly sampled geospatial data.