Practical evaluation of AI models to forecast a respiratory seasonal epidemic in Canada

Public health organizations can benefit greatly from infectious disease epidemic forecasting to guide and support interventions and policies that reduce morbidity. However, accurate and reliable epidemic forecasting remains notoriously challenging and is an evolving field in which new scientific advances continue to support forecasting efforts. Computational models that underpin artificial intelligence (AI) models are promising developments that may improve seasonal epidemic forecasting. This study aims to compare the performance of a selection of AI models previously identified as well adapted to epidemic forecasting, including machine learning (XGBoost), deep learning models (Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Transformer), time-series foundation model (Chronos), and a hybrid model: physics-informed neural networks (PINNs), against statistical models (Autoregressive Integrated Moving Average (ARIMA), Prophet) and a compartmental epidemiological model (CEM) in seasonal epidemic forecasting across Canada. The study uses weekly positivity-rate incidence of seasonal influenza A across eight Canadian provinces during the 2025/2026 season as a test case. Each model generated short-term (up to 5 weeks ahead) forecasts independently for each province from October 11, 2025, to March 28, 2026. We evaluated performance using mean absolute error (MAE) for each province. Our results show that AI models, specifically GRU, Transformer, and Chronos models, achieve lower MAE than CEM in four of eight provinces (Alberta, British Columbia, Prince Edward Island, and Saskatchewan). Meanwhile, CEM and ARIMA outperform all AI models in three provinces (Ontario, Quebec, Manitoba) and one province (Nova Scotia), respectively. Furthermore, data pooling strategies that leverage multi-province data for the target virus or multi-virus data within individual provinces greatly improve deep learning models, and Chronos performs well under zero-shot forecasting in British Columbia. In contrast, Prophet always has the highest MAE, and PINN is worse than CEMs. Overall, AI models performed comparably to, or slightly better than, CEMs. However, these modest gains may not offset their lack of interpretability, high computational costs, and reliance on the specialized expertise required for implementation. We therefore conclude that AI models are best as a complement, rather than a replacement, to CEMs, which remain crucial for interpretability and informing decision-making.

Authors

Institutions

Publication Details

Journal
medRxiv
Published
2026-11-17
DOI
https://doi.org/10.64898/2026.11.14.26368122
Primary Topic
Data-Driven Disease Surveillance
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Practical evaluation of AI models to forecast a respiratory seasonal epidemic in Canada

Warsame Yusuf, David Champredon, Monica Gabriela Cojocaru, Benjamin Longaou Benteke
2 links
medRxiv
Data-Driven Disease Surveillance
preprint

Practical evaluation of AI models to forecast a respiratory seasonal epidemic in Canada

Warsame Yusuf, David Champredon, Monica Gabriela Cojocaru, Benjamin Longaou Benteke
preprint en

Abstract

Public health organizations can benefit greatly from infectious disease epidemic forecasting to guide and support interventions and policies that reduce morbidity. However, accurate and reliable epidemic forecasting remains notoriously challenging and is an evolving field in which new scientific advances continue to support forecasting efforts. Computational models that underpin artificial intelligence (AI) models are promising developments that may improve seasonal epidemic forecasting. This study aims to compare the performance of a selection of AI models previously identified as well adapted to epidemic forecasting, including machine learning (XGBoost), deep learning models (Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Transformer), time-series foundation model (Chronos), and a hybrid model: physics-informed neural networks (PINNs), against statistical models (Autoregressive Integrated Moving Average (ARIMA), Prophet) and a compartmental epidemiological model (CEM) in seasonal epidemic forecasting across Canada. The study uses weekly positivity-rate incidence of seasonal influenza A across eight Canadian provinces during the 2025/2026 season as a test case. Each model generated short-term (up to 5 weeks ahead) forecasts independently for each province from October 11, 2025, to March 28, 2026. We evaluated performance using mean absolute error (MAE) for each province. Our results show that AI models, specifically GRU, Transformer, and Chronos models, achieve lower MAE than CEM in four of eight provinces (Alberta, British Columbia, Prince Edward Island, and Saskatchewan). Meanwhile, CEM and ARIMA outperform all AI models in three provinces (Ontario, Quebec, Manitoba) and one province (Nova Scotia), respectively. Furthermore, data pooling strategies that leverage multi-province data for the target virus or multi-virus data within individual provinces greatly improve deep learning models, and Chronos performs well under zero-shot forecasting in British Columbia. In contrast, Prophet always has the highest MAE, and PINN is worse than CEMs. Overall, AI models performed comparably to, or slightly better than, CEMs. However, these modest gains may not offset their lack of interpretability, high computational costs, and reliance on the specialized expertise required for implementation. We therefore conclude that AI models are best as a complement, rather than a replacement, to CEMs, which remain crucial for interpretability and informing decision-making.

medRxiv
Public Health Agency of Canada (CA), University of Guelph (CA)
Good health and well-being
Data-Driven Disease Surveillance
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon