G-SEIR: A Graph-Enhanced Epidemic Forecasting System for Dengue Intervention

Dengue fever poses significant public-health risks in tropical regions, with recent outbreaks in Tainan City underscoring the need for timely, fine-grained forecasting. We introduce Graph-SEIR~(G-SEIR), an end-to-end epidemic forecasting and visualization system that integrates a learnable SEIR simulator with a spatio-temporal graph neural network. The system automatically collects and processes multi-source environmental, demographic, and entomological data and generates weekly forecasts for 37 districts and 752 villages. Experiments on the 2015 and 2023 outbreaks show that G-SEIR consistently outperforms eleven baselines in Root Mean Squared Error and Mean Absolute Error. A web-based interface provides interactive, district-level risk visualization, enabling authorities to detect emerging hotspots and support early intervention. These capabilities position G-SEIR as a practical AI-driven decision-support module for smart-city governance in dengue-endemic regions, strengthening municipal capacity for proactive disease control and resource allocation.

Recommended citation: Cheng-Ru Chou, Shou-Po Tsai, Jing-Zhe Huang, Hsu-Han Tu, Pei-Xuan Li, Hsun-Ping Hsieh. G-SEIR: A Graph-Enhanced Epidemic Forecasting System for Dengue Intervention. In Proceedings of the ACM on Web Conference 2026 (WWW '26).
Download Paper | Download Slides