Fluoride surveillance in a fluoride endemic region using machine learning techniques: A case study of Vea Catchment, Upper East Region, Ghana

Abstract

This study applied machine learning techniques to monitor and predict fluoride concentrations in the Vea Catchment area, a region affected by endemic fluoride contamination. The aim was to identify high-risk areas with unsafe fluoride levels, supporting public health surveillance and water management efforts. Water quality data, including fluoride concentrations and various physico-chemical parameters, were collected from multiple sampling sites. Machine learning models, including Logistic Regression, Random Forest, and Gradient Boosting, were used for regression, classification, and spatial analysis to classify regions as safe or unsafe based on World Health Organization (WHO) fluoride guidelines. The results demonstrated the effectiveness of the models, with Random Forest and Gradient Boosting achieving high accuracy in predicting unsafe fluoride levels with precision and F1-score values of 0.88 for both models, successfully identifying high-risk areas with concentrations exceeding the WHO limit of 1.5 mg/L. Anomaly detection techniques revealed localized areas of concern. This study highlights the value of machine learning in water quality management, providing a data-driven approach to predicting fluoride contamination and informing public health interventions and water management strategies. The integration of predictive modeling with spatial analysis represents a significant advancement, offering the potential for real-time water quality monitoring in fluoride-endemic regions.