Artificial Intelligence in Disease Surveillance: Development and Validation of a Predictive Early Warning Model for Meningitis Outbreaks in Nigeria’s Meningitis Belt

Nigeria's meningitis belt, spanning the North West, North Central, and North East zones, experiences recurrent bacterial meningitis epidemics with high case fatality ratios, yet early warning and response systems remain reactive rather than predictive. This study develops and validates an artificial intelligence-based predictive early warning model for meningitis outbreaks in Nigeria's meningitis belt, integrating epidemiological, environmental, climatic, and entomological data streams. A longitudinal model development study is designed using 15 years of meningitis case notification data (2009 to 2023) from 180 local government areas across twelve northern states, combined with Saharan dust index data, Harmattan wind speed and humidity records, rainfall anomaly data, and vaccination campaign histories. Machine learning algorithms including Random Forest, Gradient Boosting, Long Short-Term Memory recurrent neural networks, and eXtreme Gradient Boosting are trained and cross-validated on an 80:20 training-test split. Model performance is evaluated using area under the receiver operating characteristic curve, sensitivity, specificity, and positive predictive value. Explainability analysis using SHapley Additive exPlanations identifies the most predictive features. Bayesian updating procedures allow the model to incorporate real-time surveillance signals. The Integrated Surveillance and Response Theory and Digital Epidemiology framework provide the conceptual basis. The original contribution is an externally validated, deployable AI prediction tool for the Nigeria Centre for Disease Control's integrated disease surveillance and response system. Keywords: artificial intelligence, meningitis, early warning, disease surveillance, Nigeria meningitis belt.