Foundation models capture human and environmental context to improve dengue forecasting
- 1University of Oxford
Abstract
Accurate and timely probabilistic dengue forecasts can support public health preparedness, but producing reliable forecasts across locations and time horizons remains challenging. We evaluated five statistical, machine-learning, deep-learning, and foundation-model approaches for forecasting monthly reported dengue cases across 5,570 Brazilian municipalities at 1-, 3-, 6- and 12-month horizons, with forecasts backtested from January 2020 to August 2025. The time-series foundation model TimesFM maintained comparatively stable probabilistic forecast performance at longer horizons, while most other models showed declining performance. We then tested whether spatial representations of human and environmental context from a Population Dynamics Foundation Model (PDFM) could improve forecasts through post-hoc residual adjustment. PDFM adjustment improved mean probabilistic forecast performance, measured by the Weighted Interval Score (WIS), in 19 of 20 model-horizon combinations, although the magnitude and frequency of improvements varied across models, horizons and municipalities. The largest improvements were observed for N-BEATS, including a reduction in mean WIS of 1.13 at the six-month horizon (95% CI [0.84, 1.48]; Diebold-Mariano p < 0.0001), with improvements also observed for XGBoost. The proportion of forecasts improved by PDFM generally increased with locally reported dengue case counts, although this relationship varied between models. Covariate selection, spatial fitting and forecast aggregation strategies showed strong model-dependent effects, while monthly model retraining provided no consistent advantage in the settings evaluated. Together, these findings highlight the potential of pretrained representations for dengue forecasting, with benefits dependent on forecasting architecture, horizon and modelling configuration.
Related
- Benchmarking time series foundation models for multi-country infectious disease forecasting: a retrospective evaluation across over 70 diseases
- Maternal–Fetal Oncology Intelligence: Self-Evolving Foundation Models and Digital Twin Ecosystems for Precision Cancer Prediction, Treatment Optimization, and Maternal–Fetal Outcome Forecasting
- Using the Haddon matrix to improve environmental road safety factors around a school zone
- Stain Normalisation Does Not Consistently Improve Matched-Location Retrieval with Pathology Foundation Models
- Proteomic Signatures Capture Environmental Exposures Across Diverse Populations
BibTeX
@misc{brittain2026foundation,
title = {{Foundation models capture human and environmental context to improve dengue forecasting}},
author = {J.-S. Brittain and C. Mills and C. Judge and L. S. Bastos and L. P. Freitas and L. M. Carvalho and E. C. Araujo and F. C. Coelho and Y. Xu and B. Reddy and R. K. Pasumarthi and T. Shekel and S. Shetty and A. C.-Y. Chou and A. Carter and A. Muslim and G. Prasad and K. Widner and M. Bharel and S. V. Scarpino and S. Flaxman and S. Bhatt and J. L. H. Tsui and M. U. G. Kraemer},
howpublished = {medRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.05.26367335},
url = {https://www.medrxiv.org/content/10.64898/2026.11.05.26367335v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Foundation models capture human and environmental context to improve dengue forecasting
AU - J.-S. Brittain
AU - C. Mills
AU - C. Judge
AU - L. S. Bastos
AU - L. P. Freitas
AU - L. M. Carvalho
AU - E. C. Araujo
AU - F. C. Coelho
AU - Y. Xu
AU - B. Reddy
AU - R. K. Pasumarthi
AU - T. Shekel
AU - S. Shetty
AU - A. C.-Y. Chou
AU - A. Carter
AU - A. Muslim
AU - G. Prasad
AU - K. Widner
AU - M. Bharel
AU - S. V. Scarpino
AU - S. Flaxman
AU - S. Bhatt
AU - J. L. H. Tsui
AU - M. U. G. Kraemer
PY - 2026
DA - 2026/11/05
PB - medRxiv
DO - 10.64898/2026.11.05.26367335
UR - https://www.medrxiv.org/content/10.64898/2026.11.05.26367335v1
AB - Accurate and timely probabilistic dengue forecasts can support public health preparedness, but producing reliable forecasts across locations and time horizons remains challenging. We evaluated five statistical, machine-learning, deep-learning, and foundation-model approaches for forecasting monthly reported dengue cases across 5,570 Brazilian municipalities at 1-, 3-, 6- and 12-month horizons, with forecasts backtested from January 2020 to August 2025. The time-series foundation model TimesFM maintained comparatively stable probabilistic forecast performance at longer horizons, while most other models showed declining performance. We then tested whether spatial representations of human and environmental context from a Population Dynamics Foundation Model (PDFM) could improve forecasts through post-hoc residual adjustment. PDFM adjustment improved mean probabilistic forecast performance, measured by the Weighted Interval Score (WIS), in 19 of 20 model-horizon combinations, although the magnitude and frequency of improvements varied across models, horizons and municipalities. The largest improvements were observed for N-BEATS, including a reduction in mean WIS of 1.13 at the six-month horizon (95% CI [0.84, 1.48]; Diebold-Mariano p < 0.0001), with improvements also observed for XGBoost. The proportion of forecasts improved by PDFM generally increased with locally reported dengue case counts, although this relationship varied between models. Covariate selection, spatial fitting and forecast aggregation strategies showed strong model-dependent effects, while monthly model retraining provided no consistent advantage in the settings evaluated. Together, these findings highlight the potential of pretrained representations for dengue forecasting, with benefits dependent on forecasting architecture, horizon and modelling configuration.
ER -