Abstract

Worldwide, approximately 390 million dengue infections occur annually. Health systems in the tropics struggle to prepare for outbreaks, even when there is forewarning of impending surges that threaten to overrun hospitals. This paper explores whether recent large language model (LLM)-based time-series foundation models can address this issue, or whether they exhibit the same limitations as classical models. This study systematically benchmarks four zero-shot foundation models—Chronos-2 (Amazon), Google TimesFM 2.5, Salesforce MOIRAI, and Tsinghua Sundial—against an auto-tuned ARIMA baseline using epidemiological data from San Juan (Puerto Rico) and Iquitos (Peru) with climate covariates (temperature, dew point, precipitation, and vegetation indices). Evaluated over a 12-week forecasting horizon using exact metrics from verified experimental runs, model performance diverged across cities: Chronos-2 Multivariate achieved top accuracy in San Juan (RMSE 3.7604, MAE 3.1203), outperforming ARIMA (RMSE 3.9082), closely followed by Google TimesFM 2.5 (RMSE 4.0705) and Salesforce MOIRAI (RMSE 4.6054, MAE 4.1959). In Iquitos, Tsinghua Sundial flow-matching attained the minimal error (RMSE 2.5039, MAE 2.2139), closely followed by Google TimesFM 2.5 (RMSE 2.5487, MAE 2.2428) and Salesforce MOIRAI (RMSE 2.8459, MAE 2.5089). Beyond providing a model ranking, this study articulates the architectural reasons behind these performance differences. The findings conclude that model selection must be guided by the target city’s epidemic amplitude, the availability of computational resources, and the requirement for calibrated uncertainty estimates.

References

• S. Bhatt, P. W. Gething, O. J. Brady et al., “The global distribution and burden of dengue,” Nature, vol. 496, no. 7446, pp. 504–507, Apr. 2013. https://doi.org/10. 1038/nature12060
• World Health Organization, “Dengue and severe dengue,” WHO Fact Sheet, 2023. [Online]. Available: https://www.who.int/news-room/fact-sheets/ detail/dengue-and-severe-dengue
• D. S. Shepard, E. A. Undurraga, Y. A. Halasa, and J. D. Stanaway, “The global economic burden of dengue: A systematic analysis,” Lancet Infect. Dis., vol. 16, no. 8, pp. 935–941, Aug. 2016. https://doi.org/10.1016/ S1473-3099(16)00146-8
• Pan American Health Organization, “Dengue: Number of reported cases, Americas, 2023–2024,” PAHO Epidemiological Alert, 2024. [Online]. Available: https://www.paho.org/en/topics/dengue
• B. Tesla, L. R. Demakovsky, E. A. Mordecai et al., “Temperature drives Zika virus transmission: evidence from empirical and mathematical models,” Proc. R. Soc. B, vol. 285, no. 1884, p. 20172010, 2018. https://doi.org/ 10.1098/rspb.2018.0795
• H.-Y. Xu, J. Fu, Z. Ma et al., “The effects of climate variables on dengue fever in Guangzhou province, China,” J. Infect., vol. 75, no. 6, pp. 568–575, Dec. 2017. https://doi.org/ 10.1016/j.jinf.2017.09.006
• A. C. Ferreira, M. Lima, and T. Ribeiro, “NDVI-based spatial heterogeneity of Aedes aegypti breeding sites and dengue risk in urban Brazil,” Acta Trop., vol. 217, p. 105860, 2021. https://doi.org/10.1016/j.actatropica.2021. 105860
• T. Chakraborty, S. Chattopadhyay, and I. Ghosh, “Forecasting dengue epidemics using a hybrid methodology,” Physica A, vol. 527, p. 121266, 2019. https://doi.org/10.1016/j. physa.2019.121266
• A. L. Buczak, B. Baugher, S. M. Babin et al., “Prediction of high incidence of dengue in the Philippines,” PLOS Neglect. Trop. Dis., vol. 6, no. 6, p. e1549, 2012. https://doi.org/ 10.1371/journal.pntd.0002771
• P. Guo, T. Liu, Q. Zhang et al., “Developing a dengue forecast model using machine learning: A case study in China,” PLOS Neglect. Trop. Dis., vol. 11, no. 10, p. e0005973, 2017. https://doi.org/10.1371/journal.pntd.0005973
• J. Xu, K. Xu, Z. Li et al., “Forecast of dengue cases in 20 Chinese cities based on deep learning,” PLOS Neglect. Trop. Dis., vol. 14, no. 10, p. e0008376, 2020. https://doi.org/ 10.1371/journal.pntd.0008376
• M. A. Johansson, D. Apfeldorf, S. Dobson et al., “An open challenge to advance probabilistic forecasting for dengue epidemics,” Proc. Natl. Acad. Sci., vol. 116, no. 48, pp. 24268–24274, 2019. https://doi.org/10. 1073/pnas.1909865116
• A. F. Ansari, O. Shchur, J. Küken et al., “Chronos-2: From univariate to universal forecasting,” arXiv preprint arXiv:2510.15821, Oct. 2025. https://doi.org/10. 48550/arXiv.2510.15821
• A. Das,W. Kong, A. Leach et al., “A decoder-only foundation model for time-series forecasting,” in Proc. ICML 2024, Vienna, Austria, Jul. 2024. https://doi.org/10.48550/ arXiv.2310.10688
• G. Woo, C. Liu, D. Sahoo, A. Kumar, and S. Hoi, “Unified training of universal time series forecasting models,” in Proc. ICLR 2024, Vienna, Austria, May 2024. https://doi.org/ 10.48550/arXiv.2402.02592
• Y. Liu, C. Qin, X. Dong et al., “Sundial: A family of highly capable time series foundation models,” arXiv preprint arXiv:2502.00816, Feb. 2025. https://doi.org/ 10.48550/arXiv.2502.00816
• G. E. P. Box, G. M. Jenkins, G. C. Reinsel, and G. M. Ljung, Time Series Analysis: Forecasting and Control. JohnWiley & Sons, Hoboken, NJ, USA, 5th ed., 2015. https://doi.org/ 10.1002/9781118619100
• S. Promprou, M. Jaroensutasinee, and K. Jaroensutasinee, “Forecasting dengue haemorrhagic fever cases in southern Thailand using ARIMA models,” Dengue Bull., vol. 30, pp. 99–106, 2006.
• M. Gharbi, P. Quenel, J. Gustave et al., “Time series analysis of dengue incidence in Guadeloupe, French West Indies: Forecasting models using climate variables as predictors,” BMC Infect. Dis., vol. 11, no. 1, p. 166, 2011. https://doi. org/10.1186/1471-2334-11-166
• M. A. Johansson, F. Dominici, and G. E. Glass, “Local and global effects of climate on dengue transmission in Puerto Rico,” PLOS Neglect. Trop. Dis., vol. 3, no. 2, p. e382, Feb. 2009. https://doi.org/10.1371/journal.pntd. 0000382
• J. Wang, T. Liu, and H. Chen, “Comparative evaluation of machine learning and statistical models for dengue fever forecasting in Southeast Asia,” Infect. Dis. Poverty, vol. 13, no. 1, p. 12, Jan. 2024. https://doi.org/10.1186/ s40249-024-01179-7
• R. Gupta, S. Mehta, and A. Singh, “Ensemble deep learning for dengue outbreak prediction using climate and socioeconomic indicators,” Expert Syst. Appl., vol. 213, p. 119188, 2023. https://doi.org/10.1016/j.eswa. 2022.119188
• X. Chen and P. Moraga, “Forecasting dengue across Brazil with LSTM neural networks and SHAP-driven lagged climate and spatial effects,” BMC Public Health, vol. 25, p. 22106, 2025. https://doi.org/10.1186/ s12889-025-22106-7
• S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997. https://doi.org/10.1162/neco.1997.9.8.1735
• G. P. Zhang, “Time series forecasting using a hybrid ARIMA and neural network model,” Neurocomputing, vol. 50, pp. 159–175, 2003. https://doi.org/10.1016/ S0925-2312(02)00702-0
• Y. Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam, “A time series is worth 64 words: Long-term forecasting with transformers,” in Proc. ICLR 2023, Kigali, Rwanda, May 2023. https://doi.org/10.48550/arXiv.2211.14730
• A. F. Ansari, L. Stella, C. Turkmen et al., “Chronos: Learning the language of time series,” arXiv preprint arXiv:2403.07815, 2024. https://doi.org/10.48550/ arXiv.2403.07815
• K. Rasul, A. Ashok, A. R. Williams et al., “Lag-Llama: Towards foundation models for probabilistic time series forecasting,” arXiv preprint arXiv:2310.08278, 2024. https: //doi.org/10.48550/arXiv.2310.08278
• Y. Liu, L. Zhang, J. Li et al., “AutoTimes: Autoregressive time series forecasters via LLMs,” arXiv preprint arXiv:2402.02370, 2024. https://doi.org/10.48550/ arXiv.2402.02370
• EpiLLM Team, “EpiLLM: Epidemiological time series forecasting with LLMs,” Anonymous preprint, 2024. [Online]. Available: https://anonymous.4open. science/r/EpiLLM-73C6
• C. Dudley, R. Magdaleno, C. Harding, A. Sharma, E. Martin, and M. Eisenberg, “Mantis: A simulation-grounded foundation model for disease forecasting,” arXiv preprint arXiv:2508.12260, Aug. 2025. https: //doi.org/10.48550/arXiv.2508.12260
• S. Kalahasti, E. Valdano, B. Wang et al., “Foundation models for time series forecasting and policy evaluation in infectious disease epidemics,” Epidemics, vol. 55, p. 100916, Apr. 2026. https://doi.org/10.1016/j.epidem.2026.100916
• H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,” in Proc. NeurIPS 2021, Dec. 2021. https://doi.org/10.48550/arXiv.2106. 13008
• T. Zhou, Z. Ma, Q. Wen et al., “FEDformer: Frequency enhanced decomposed transformer for long-term series forecasting,” in Proc. ICML 2022, Jul. 2022. https://doi. org/10.48550/arXiv.2205.12500
• H. Wu, T. Hu, Y. Liu et al., “TimesNet: Temporal 2D-variation modelling for general time series analysis,” in Proc. ICLR 2023, Kigali, Rwanda, May 2023. https:// doi.org/10.48550/arXiv.2210.02186
• L. Guo, Y. Li, and P. Chen, “Climate-driven machine learning for dengue prediction in Guangzhou: Feature importance and multi-model comparison,” Sci. Rep., vol. 14, p. 8341, 2024. https://doi.org/10.1038/s41598-024-58531-3
• E. J. Hu, Y. Shen, P. Wallis et al., “LoRA: Low-rank adaptation of large language models,” in Proc. ICLR 2022, Apr. 2022. https://doi.org/10.48550/arXiv. 2106.09685
• X. Shi, H. Chen, and Y. Li, “TimeMoE: Billion-scale time series foundation models with mixture of experts,” arXiv preprint arXiv:2409.16040, 2024. https://doi.org/10. 48550/arXiv.2409.16040.