From data to decision: leveraging machine learning and water quality index for groundwater quality evaluation

Abstract

Groundwater quality is critical for sustainable development, serving as a primary source of drinking water and irrigation. The present study employs the machine learning (ML) models to evaluate the water quality index (WQI) in order to enhance the groundwater quality assessment. Forty groundwater samples were collected from six diverse locations and analyzed for seven physicochemical parameters, including pH, Turbidity, CO₂, Chloride, Alkalinity, TDS, and Fe. To improve model generalizability, data augmentation techniques, Gaussian noise and interpolation, expanded the dataset to 120 samples. WQI was computed using the Canadian Council of Ministers of the Environment (CCME) method. Six ML models were employed for predictive analysis and evaluated based on R2, RMSE, and MAE. The results revealed significant contamination, with 25% of samples exceeding acceptable limits for total dissolved solids (TDS), while iron levels averaged 3.01 mg/L, ten times higher than the WHO guideline of 0.3 mg/L. WQI values ranged from 45.89 to 100, classifying most samples as "Fair to Good" but identifying critical degradation in specific areas. Among the six ML models tested, XG-Boost outperformed the others, achieving the highest predictive accuracy (R2 = 0.97, RMSE = 1.72, MAE = 1.38). These findings highlight substantial groundwater contamination risks, particularly from iron and turbidity. This research demonstrates the effectiveness of ML in groundwater quality assessment, providing a scalable decision-support framework for environmental management and policymaking in resource-limited regions.