Combining clustering and ensemble learning for groundwater quality monitoring: a data-driven framework for sustainable water management.

This groundwater quality assessment study for the state of Punjab, India, utilized six Water Quality Index (WQI) models, i.e., NSF-AM, NSF-GM, CCME, Horton, West Java, and GPI for potability assessment via machine learning (ML) classifiers. The results of the study manifested poor groundwater quality in many regions of the state that fall below potability standards (WHO and BIS). CCME WQI classified the state's groundwater as poor to marginal, rendering it unsuitable for human consumption. The disparities observed among WQI models highlighted differences in parameter selection, weight assignment, and aggregation techniques, emphasizing the need for a customized WQI framework for the Indian subcontinent for more accurate and robust groundwater quality assessment. K-means clustering, employed as a preprocessing step for improving classification accuracy, grouped data into two distinct clusters (validated by silhouette scores = 0.927 and Calinski-Harabasz index = 129.21), revealing contamination sources' patterns, feature refinement, and enhancement. Further, application and performance analysis of ML classifiers integrated with K-means clustering analysis identified Ensemble Hard Voting (EHV) and Ensemble Soft Voting (ESV) as top performers for groundwater quality classification. The GPI WQI combined with ESV achieved Accuracy = 99.13%, Precision = 100%, Recall = 99.03%, F1-score = 99.51%, Specificity = 100%, MCC = 0.95, Log Loss = 0.11, and AUC = 100% while maintaining moderate model and computational complexity (t_predict = 0.0095 s), underscores the efficiency and suitability of GPI and ESV blend for real-time water quality monitoring systems. The presented data-driven holistic framework highlights the capability of ML-driven groundwater assessment as a decision-support tool for resource-constrained regions, facilitating policy interventions and promoting sustainable water management practices by leveraging its accurate classification and real-time assessment capabilities.