An innovative combination of extra trees within adaboost for accurate prediction of agricultural water quality indices

• All models' hyperparameters were adjusted using the grid search and Bayesian optimizer algorithms. • The accuracy of agricultural water quality predictions was enhanced by using extra trees as the base estimator for adaboost. • The optimal adaboost-extra trees hybrid model had the best performance in predicting SAR and PS indices. • The performance of the developed models was the same in determining the most important influencing variable. The management of water quality plays a pivotal role in fostering sustainable development, especially in areas where groundwater serves as a vital resource for agricultural purposes. This study investigates methods to predict water quality indicators like potential salinity (PS) and sodium adsorption ratio (SAR) using data-driven techniques. It explores the AdaBoost algorithm and a hybrid model that combines AdaBoost with Extra Trees (ET). The research centers on the Sarayan region in southern Khorasan, Iran, involving the collection and analysis of groundwater quality data, which includes both physical and chemical parameters, spanning a period of four years. Pearson correlation analysis is utilized in this study to determine the critical input variables for predicting Sodium Adsorption Ratio (SAR) and Potential Salinity (PS). Significant performance improvements are demonstrated through the optimization of AdaBoost and the hybrid AdaBoost-Extra Trees (ET) models, achieved via grid search and Bayesian optimization. The results indicate that the hybrid model surpasses AdaBoost in performance. Specifically, the optimized AdaBoost model achieves a Test MSE of 3.57 and a Test R² of 0.87 for SAR prediction, whereas the hybrid model obtains a significantly better Test MSE of 0.87 and Test R² of 0.97. For the prediction of PS, the optimized AdaBoost model yields a Test MSE of 0.223 and a Test R² of 0.976, while the hybrid model delivers a significantly improved Test MSE of 0.018 and Test R² of 0.998. Feature importance analysis identifies critical patterns in the relevance of input parameters. This study underscores the efficacy of hybrid data-driven models in water quality prediction, showcasing their superior accuracy.