Groundwater fluoride prediction modeling using physicochemical parameters in Punjab, India: a machine-learning approach

• Published in: Frontiers in Soil Science (Online) 2024-07, Vol.4
• Main Authors: Kerketta, Anjali, Kapoor, Harmanpreet Singh, Sahoo, Prafulla Kumar
• Format: Article
• Language: English
• Subjects: Extreme Learning Machine; groundwater fluoride; machine-learning; physicochemical parameters; prediction modeling; relative importance of variables
• ISSN: 2673-8619; 2673-8619
• DOI: 10.3389/fsoil.2024.1407502

IntroductionRising fluoride levels in groundwater resources have become a worldwide concern, presenting a significant challenge to the safe utilization of water resources and posing potential risks to human well-being. Elevated fluoride and its vast spatial variability have been documented across different districts of Punjab, India, and it is, therefore, imperative to predict the fluoride levels for efficient groundwater resources planning and management.MethodsIn this study, five different models, Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boosting (Xgboost), Extreme Learning Machine (ELM), and Multilayer Perceptron (MLP), are proposed to predict groundwater fluoride using the physicochemical parameters and sampling depth as predictor variables. The performance of these five models was evaluated using the coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE).Results and discussionELM outperformed the remaining four models, thus exhibiting a strong predictive power. The R2, MAE, and RMSE values for ELM at the training and testing stages were 0.85, 0.46, 0.36 and, 0.95, 0.31, and 0.33, respectively, while other models yielded inferior results. Based on the relative importance scores, total dissolved solids (TDS), electrical conductivity (EC), sodium (Na+), chloride (Cl−), and calcium (Ca2+) contributed significantly to model performance. High variability in the target (fluoride) and predictor variables might have led to the poor performance of the models, implying the need for better data pre-processing techniques to improve data quality. Although ELM showed satisfactory results, it can be considered a promising model for predicting groundwater quality.