Kan AI forudsige spredningen af en smitsom sygdom i en by udelukkende ved hjælp af anonymiserede mobilitetsdata ?

Folkens sundhedsmyndigheder stoler i stigende grad på data-drevne modeller til at forudse sygdomsudbrud, men mange kræver følsomme personoplysninger eller komplekse simuleringer. En nylig AI-evne indebærer at forudsige spredning af smitsomme sygdomme ved hjælp af anonymiserede datasæt over menneskers bevægelsesmønstre. AI’en skal tage højde for variationer i adfærd, befolkningstæthed og miljømæssige faktorer for at producere handlingsrettede, yderst præcise forudsigelser.

Background

Public health officials increasingly rely on data-driven models to anticipate disease outbreaks, but many require sensitive personal data or complex simulations. A recent AI capability involves forecasting infectious disease spread using anonymized datasets of human movement patterns. The AI must account for variations in behavior, population density, and environmental factors to produce actionable, highly accurate predictions.

AI systems can now estimate disease spread from anonymized mobility data by treating trips as vectors for transmission and running Monte Carlo simulations over contact networks inferred from location traces. Models such as Epifcast, Epigram, and deep-learning approaches that combine graph neural networks with mobility embeddings report median absolute errors around 3–8 % for weekly incidence forecasts in cities like Boston and Singapore, outperforming gravity and radiation baselines. These methods typically rely on aggregated mobile-phone location pings rather than raw trajectories, applying differential privacy or k-anonymity to preserve anonymity while retaining coarse mobility patterns.

— Enriched May 13, 2026 · Source: Nature Communications