Groundwater is key to ecosystem health, food production and interlinked hydrological processes in dryland regions under a changing climate, therefore assessment of groundwater quality becomes imperative. In this study, performance of different Machine Learning Models (MLMs) to predict the Entropy Water Quality Index (EWQI) of groundwater of a vast dryland area experiencing frequent & prolonged drought is elaborated. In addition, the scheme for selecting training and testing datasets to achieve higher accuracy in model outputs is highlighted. An extensive set of hydrochemical data (over 800) covering different seasons was chosen for testing the performance of four MLMs viz., Multi-Linear Regression (MLR), Lasso Regression (LaR), eXtreme Gradient Boosting (XGBoost), and Random Forest (RF). Analyzing the R2 (0.98 to 1.0), Mean Absolute Error (MAE) (3.9 × 10− 9 to 2.83), Mean Squared Error (MSE) (6.5 × 10− 17 to 13.87) and Root Mean Square Error (RMSE) (8.07 × 10− 9 to 3.72), the MLR and LaR models exhibited higher accuracy than RF and XGBoost. Analysis of EWQI trends also reveals a decline in water quality from 2019 to 2022. The percentage of water samples classified as “excellent” has decreased, shifting towards “good” and “poor” classes. The percentage of water samples in the “very poor” and “unsuitable” classes has increased, indicating deterioration in groundwater quality. This study provides an optimum methodology for achieving robust and accurate Machine Learning Model (MLM) outputs for the dryland conditions of the present study area.
Kumari et al. (Sun,) studied this question.