Machine learning-based multivariate risk stratification framework for assessing the combined burden of occupational, infectious, and non-communicable diseases in Bagega (Zamfara) and Shiroro (Niger), Nigeria

The case study describes a machine learning-driven stratification of syndemic health risks in resource-limited artisanal mining populations, where occupational risks, infectious diseases, and non-communicable conditions overlap under poor ecological and infrastructural conditions. Trained on 3,000 adult supervised data points in Northern Nigeria, the XGBoost classifier, optimised with grid search and explained with SHAP, proved superior to logistic regression and neural networks, achieving an AUROC of 0.93 and an F1- score of 0.86. A total of twenty-three percent of participants were high-risk, especially older males with an experience of over a decade of ore crushing, limited access to clean water, pit latrines or gaps in the open, and household overcrowding. The prevalence of malaria stood at 70.5% and hypertension, chronic cough, and tuberculosis-like symptoms were prevalent. More than 30% had two or more comorbidities, and 69.7% resided within 500 meters of mining areas. Only 27.5 per cent of them used some PPE. Bagega presented 50% more high-risk cases compared to Shiroro. The Proportionate-reduction-in-error (PRE) feature attribution analysis yielded an affirmative result, indicating that structural exposures, rather than formal diagnoses, were the prevailing risk drivers. The framework provides a scalable method that is interpretable and enables precision-specific surveillance and adaptation targeting of interventions in high-weighted environments where data is immediately limited.