<p>Biochemical oxygen demand (BOD) is crucial for assessing the health of aquatic ecosystems and essential for implementing efficient water management techniques to safeguard water resources and promote sustainable development. Alternative approaches, like machine learning models, have been suggested for BOD prediction since conventional methods are time-consuming and cost-intensive. This study proposed and examined the ability of the supervised machine learning method, extreme gradient boosting (XGBoost), to evaluate BOD using various water quality parameters as covariates on the mainstream of River Ganga, India, for limited water quality dataset. The proposed model was also compared with two other models: adaptive boosting (AdaBoost) and artificial neural network (ANN). Feature selection analysis involved input variables like pH, dissolved oxygen (DO), electrical conductivity (EC), nitrate (NO₃⁻), fecal coliform (FC), total coliform (TC), and the states that the river runs through for developing the models. Overall findings of the study revealed that the XGBoost model showed good prediction since it provided the most precise BOD measurement, had the lowest RMSE (0.5184), and had a prediction success rate of 79.85% compared to the other models. These models may help water organizations establish a decision-making framework that will assist them in addressing water quality deterioration warnings in rivers like the Ganga.</p>