Spatial Modeling of Environmental and Clinical Drivers of Meningitis Transmission in the Upper West Region of Ghana Using Geospatial Machine Learning

Abstract Background Meningitis remains a critical public health threat in the "Meningitis Belt" of sub-Saharan Africa, necessitating high-resolution predictive tools for targeted intervention. This study model environmental and clinical drivers of bacterial meningitis transmission and map population exposure risk in the Upper West Region of Ghana using machine learning. Methods PCR-confirmed case data (2018–2019) were integrated with multi-source geospatial predictors at a 100-meter resolution. Six models (Logistic Regression, Artificial Neural Networks, Support Vector Machines, Random Forest, Gradient Boosting Machine, and Extreme Gradient Boosting) were trained using repeated 5-fold cross-validation. A population-prevalence adjustment was applied to exposure at 1 km² resolution. Results Extreme Gradient Boosting showed the best clinical performance (F1-score: 0.478; sensitivity: 0.407), while Random Forest achieved the highest accuracy (83.5%). Annual rainfall (OR = 0.32, p < 0.0001), waterbodies (OR = 0.65, p = 0.002), and surface temperatures (OR = 0.55, p = 0.001) were significant environmental factors that reduced infection risk. Conversely, neck stiffness (OR = 1.60, p = 0.002) and night-time light (OR = 1.49, p = 0.041) were associated with increased odds of infection. Nandom, Lawra, and Jirapa Lambussie emerged as the primary geographical hotspots. Conclusion This study highlights the critical interplay between environmental factors and clinical indicators in shaping the risk of bacterial meningitis. Integrating geospatial environmental data with machine learning offers a powerful framework for targeted surveillance and early warning systems in endemic regions.