A Scoping Review of Artificial Intelligence and Machine Learning for Spatio-temporal Infectious Disease Forecasting

Abstract Infectious diseases spread across space and time, yet most artificial intelligence (AI) and machine learning (ML) models treat forecasting as a purely temporal problem. We conducted a scoping review of AI/ML methods for spatio-temporal infectious disease forecasting, searching five databases on 20th November 2025. We included 93 studies published between 2018 and 2026, spanning 14 infectious diseases. Recurrent neural networks (46%) and graph neural networks (39%) dominated, with hybrid mechanistic-ML models growing steadily after 2022. Over 30 distinct error metrics were used, with no shared benchmarks, making cross-study comparison infeasible. Uncertainty was quantified in only 19% of studies, external validation in 13%, and 71% did not share code openly. Forecasts were concentrated in middle- and high-income countries, and only one study conducted fairness analysis. Addressing these gaps in evaluation standardisation, transparency, and equity is essential for AI/ML forecasting to inform public health policy.