DengueGNN: Graph-based deep learning for modeling disease spread dynamics and prediction

Comprehensive Summary

The study introduces DengueGNN, a dynamic spatio‑temporal graph neural network designed to predict dengue fever outbreaks by integrating human mobility, environmental factors, historical case data, and spatial relationships between regions. The model constructs a time‑varying graph where nodes represent geographic areas and edges reflect both mobility flows and geographic adjacency. DengueGNN combines graph convolution, attention‑enhanced LSTMs, and feature‑fusion modules to capture complex epidemic dynamics. Evaluated using the OpenDengue dataset, the model outperforms traditional statistical methods and deep learning baselines across 1‑, 2‑, and 4‑week forecasting horizons.

Outcomes and Implications

DengueGNN demonstrates that incorporating dynamic mobility and environmental data into graph‑based deep learning significantly improves outbreak prediction accuracy. This approach can help public health agencies identify high‑risk regions earlier, allocate resources more effectively, and design targeted interventions. The model’s explainability tools also support transparent decision‑making, making it a promising framework for real‑world epidemic surveillance and early‑warning systems.