Classifying arsenic-contaminated waters in Tarkwa: a machine learning approach

• Published in: Sustainable water resources management 2024-04, Vol.10 (2), p.55, Article 55
• Main Authors: Ayisha, Mohammed, Nkoom, Matthew, Doke, Dzigbodi Adzo
• Format: Article
• Language: English
• Subjects: Algorithms; Arsenic; Classification; Contamination; Decision trees; Developing countries; Development Economics; Dissolved solids; Drinking water; Earth and Environmental Science; Earth Sciences; Electrical conductivity; Electrical resistivity; Global health; Hydrogeology; Hydrology/Water Resources; LDCs; Learning algorithms; Machine learning; Mathematical models; Monitoring; Original Article; Parameters; Public health; Sensitivity analysis; Sustainable Development; Total dissolved solids; Turbidity; Water monitoring; Water Policy/Water Governance/Water Management
• ISSN: 2363-5037; 2363-5045
• DOI: 10.1007/s40899-024-01042-1

Access to clean and safe drinking water is key to the improvement of social lives in most developing countries. Due to its hazardous nature and detrimental effects on human health, increased quantities of arsenic in water bodies have been a growing global health concern in recent years. In Ghana, elevated arsenic concentration is reported in some waters in Tarkwa. However, constant monitoring of arsenic concentrations in these water sources are inhibited by the associated huge expenses. To facilitate early detection, this study aimed at developing efficient machine learning models for classifying high, medium and low levels of arsenic contamination using physical water parameters, such as total dissolved solids, pH, electrical conductivity and turbidity. These parameters were selected, because they are relatively inexpensive to measure, their data were available and they may influence the concentration of arsenic in the water. Thus, three machine learning models, namely, extra trees, random forest and decision tree, were developed and assessed using evaluation metrics, such as accuracy, precision and sensitivity. The evaluation results justified the superiority of the extra trees and random forest models over decision tree. However, all developed machine learning models generally gave remarkable performance when classifying waters with high and low levels of arsenic contamination. Moreover, the variable importance analysis revealed that pH had the strongest influence in classifying arsenic contaminated waters followed by electrical conductivity. The outcome of the study has revealed the potency of machine learning algorithms in assisting water monitoring practitioners for monitoring arsenic concentration in water sources.