Soft computing approaches for predicting boron contamination in arid sandstone groundwater

Groundwater is an important freshwater resource, accounting for 36% and 42% of global drinking and agricultural water use. The current study aims to develop a novel machine-learning model for predicting boron contamination in sandstone groundwater from southern Saudi Arabia. The dataset for model development incorporates physical and chemical parameters, ions composition and boron data. Five machine learning models were evaluated, including support vector regression (SVR), Gaussian process regression (GPR), decision tree (DT), random forest (RF), and artificial neural networks (ANN) for predicting boron concentration in tested groundwater using two data-splitting scenarios and six input data combinations. The results showed that the SVR model utilizing a combination of TDS, EC, K, Na, SO4, Mg, Ca, and Cl as input parameters (Model A) and a 70:30 data split (Scenario A) demonstrated superior performance and generalization capability, achieving an R2, RMSE, MAE, NSE and WI of 0.7382, 0.0753, 0.0263, 0.7347, and 0.9181 respectively in testing set evaluations. This study showcases the capacity of machine learning models to accurately predict boron contamination in groundwater systems using water quality data, offering cost-effective solutions for groundwater monitoring and management. Future research should be focused on advanced models and extended data collection across several regions to enhance the model’s robustness and applicability.