Automated Water Quality Index Classification via Gradient Boosting: Identifying BOD, Fecal Coliform and DO as Key Predictors

DOI

https://www.doi.org/10.59256/indjcst.20260503102

Published In

Volume 5, Issue 3 (2026)

Published Online

September-December 2026

Pages

859-867

Article Type

Original Article

Check for updates

Water quality monitoring is critical for public health and environmental sustainability, particularly in developing nations where freshwater pollution escalates rapidly. The Water Quality Index (WQI) consolidates multiple physico-chemical and microbiological parameters into a single interpretable value, but computing it across national networks by laboratory methods alone is slow. This paper proposes an automated WQI classification framework based on Extreme Gradient Boosting (XGBoost), which learns to assign WQI classes calculated from measured parameters; it is not a forecasting model. From the CPCB Indian water quality dataset (1,991 records, 2003–2014), 1,578 records with complete measurements of seven parameters were retained: temperature, pH, Dissolved Oxygen (DO), Biological Oxygen Demand (BOD), Electrical Conductivity (EC), Nitrate-Nitrite and Fecal Coliform (FC). A weighted arithmetic WQI computed from six of them (temperature excluded) was classified as Good, Poor or Very Poor. On a stratified held-out test set of 316 records, XGBoost 1.7.5 achieved 93.0% accuracy, 92.7% macro precision, 93.1% macro recall, 92.9% macro F1-score and a macro-average AUC of 0.993, the highest accuracy, recall, F1-score and AUC of seven classifiers, but only marginally ahead of Random Forest. BOD was the most influential predictor (0.426 of total gain), followed by FC and DO. Because the labels are derived from the inputs, the results reflect reproduction of the WQI rule, not independent prediction.