Prediction of fluoride contamination in groundwater of Nigeria using a weighted ensemble machine-learning framework

Fluoride (F^-) contamination of groundwater remains a significant health risk in Nigeria, affecting rural and urban communities. This study examines the spatial distribution and factors influencing F^- levels using a weighted ensemble machine learning model to forecast high-risk areas and estimate exposed populations. The primary factor controlling geogenic F^- is the aquifer lithology, with the highest F^- levels found in groundwater from crystalline basement complexes and sedimentary sandstones. The problem is exacerbated in dry regions, including semiarid and arid zones, where evaporation rates increase, concentrating solutes in the water. Fluoride levels exceed World Health Organization (WHO) standards at two primary geological locations: the sedimentary aquifers of the Sokoto and Middle Niger Basins, and the Precambrian crystalline basement aquifers across northern Nigeria. Land-use and land-cover conditions (croplands, urban areas, and bare soils) also significantly affect groundwater hydrochemistry. Anthropogenic inputs (agriculture and urbanization) may contribute to localized F^- mobilization. Feature importance analysis identified well depth as the most influential predictor (20.2%), followed by HCO₃^- (16.6%), Cl^- (13.8%), total cations (13.3%), and total anions (12.7%), confirming the dominance of hydrogeological and geochemical controls. The weighted ensemble model, combining XGBoost, Random Forest, Extra Trees, and Histogram-based Gradient Boosting, achieved an accuracy of 83.74% with a Gini index of 0.71 (AUC-ROC = 0.85). Among individual base learners, Extra Trees achieved the highest Gini index (0.72). The model demonstrated high specificity (89.16%) and negative predictive value (87.06%), supporting its usefulness for identifying likely safe groundwater sources. Although Extra Trees achieved the highest individual Gini index, the weighted ensemble provided balanced overall performance across multiple evaluation metrics. The analysis reveals that more than 19,443 individuals, including 9951 infants and 7795 children, are at risk, particularly in the northern regions of Nigeria. The study highlights public health concerns, including the risk of dental and skeletal fluorosis, among communities relying on untreated groundwater. Both anthropogenic and geogenic processes jointly govern F^- distribution. However, the resolution and geographic density of the input data limit its accuracy. We recommend using sophisticated methods and higher-resolution geochemical data to improve risk estimates in Nigeria and other comparable African regions.