TREA-Net: A Transferable Residual Epidemiological Adaptation Network for Dengue Incidence Forecasting

2026-07-29

AI Summary

This study addresses the challenge of scarce historical data in target regions for dengue forecasting by proposing a transferable learning framework that integrates an environmental-driven SIR model with a neural time-series backbone. The core innovation lies in a lightweight, node-invariant gated residual adaptation module, which requires learning only two global parameters to effectively transfer knowledge and capture local epidemic dynamics. Coupled with conformal prediction, the approach significantly outperforms baseline methods in 9 out of 10 transfer scenarios. When integrated with TiRex, it achieves the lowest mean absolute error and, in Mexico, reduces the 8-week prediction interval width by 29.6% while maintaining empirical coverage.

Abstract

Accurate multi-week dengue forecasting supports timely vector-control interventions, outbreak preparedness, and healthcare resource allocation. However, newly established surveillance systems often lack the historical data needed to train reliable neural forecasting models. Although pretrained time-series models offer promising zero-shot forecasts, their cross-domain training may not capture local epidemiological dynamics. We propose TREA-Net, a Transferable Residual Epidemiological Adaptation Network for dengue forecasting under limited data. TREA-Net augments neural forecasting backbones with projections from an Environmental Time-Series Susceptible-Infected-Recovered model and learns a lightweight gated residual correction transferable from data-rich to data-scarce regions. Its node-invariant design accommodates surveillance systems with different numbers of locations, while target adaptation requires learning only two global parameters. We transfer knowledge from long-running dengue surveillance in Colombia and Nicaragua to 8-week-ahead forecasting in Mexico and Malaysia using only 78 or 104 weeks of target data. Across five neural backbones and ten transfer settings, TREA-Net improves the corresponding backbone in 9 out of 10 settings, with statistically significant gains. When integrated with TiRex, a foundation model for forecasting, it achieves the lowest mean absolute error across all target datasets. Conformal prediction further maintains empirical coverage while reducing 8-week prediction-interval width by 29.6% in Mexico. These results demonstrate TREA-Net's potential as a lightweight and portable early-warning framework for health agencies with limited surveillance data.

Problem

• dengue forecasting
• limited data
• surveillance systems
• epidemiological dynamics
• transfer learning

Innovation

• transfer learning
• epidemiological modeling
• residual adaptation
• dengue forecasting
• conformal prediction