ABSTRACT Four-panel flow diagram showing the study workflow: Béchar aquifer dataset input with 621 groundwater samples and 11 physicochemical parameters; seven supervised machine learning models benchmarked using Bayesian hyperparameter optimisation and five-fold cross-validation; ANN achieving 99.47% accuracy and SVM 97.88%; deployment as a MATLAB-based groundwater quality prediction GUI tool. Accurate prediction of groundwater quality is essential for environmental monitoring and public health protection, particularly in arid regions such as Béchar in southwest Algeria. This study applied a root mean square–based water quality index (RMS–WQI) to evaluate groundwater quality using 621 samples characterized by physicochemical parameters, including pH, electrical conductivity, total dissolved solids, major cations, major anions, and nitrate. Seven supervised machine learning algorithms K-nearest neighbors, artificial neural network (ANN), support vector machine (SVM), ensemble trees (EN), discriminant analysis, Naïve Bayes, and decision trees were trained and optimized in MATLAB using the Classification Learner Toolbox with Bayesian optimization and fivefold cross-validation. Among the tested models, ANN and SVM achieved the highest predictive performance, with accuracies of 99.47 and 97.88%, respectively, along with superior precision, recall, F1-score, and Cohen's Kappa values, indicating strong agreement with observed RMS–WQI classes. Compared to conventional RMS–WQI assessment and previously reported nonoptimized models, the proposed framework demonstrates improved classification accuracy and robustness. Additionally, an operational graphical user interface was developed to facilitate rapid groundwater quality estimation using routine measurements. The findings highlight the effectiveness of optimized ANN and SVM models as reliable decision support tools for groundwater quality management in data-scarce arid environments.
Hussein et al. (Thu,) studied this question.