Water quality assessment and geospatial techniques for the delineation of surface water potential zones: A data-driven approach using machine learning models

In recent past, water demand for rising rivers like the Brahmani Basin in Odisha would be directly impacted by the threat to the natural water supply posed by the growing climate change and the conversion of green spaces into urban areas. As a result, the growing demand and decreasing water supply led to scientific research into surface water availability and sustainable management as a substitute. The model was tested with 15 physicochemical parameters which were collected from several monitoring locations and analysed using standard methods for a duration of 2019-2025 in pre - monsoon season, that represents various substances present in water samples and their concentrations. The dataset used in the study, comprised of seven locations, and the model development involves several stages, including data pre-processing. The study employed a dataset consisting of seven locations, and the model generation procedure contains multiple stages, one of which is data pre-processing. We also, employed a Geographical Information System (GIS), integrated environment coupled with a multi-criteria decision-making (MCDM) methodology, is employed for managing water resources and analysing spatial data. In the present investigation, the results of pH, reported to be slightly alkaline in nature. The temperature varies between 25-30°C. The sampling period determines how much the oxygen saturation increases; the dry season records the lowest number. The hydrogeochemical trend observed in the study area showed is Ca 2 + > Mg 2+ > Na 2+ > K + (cations), while anions are arranged as SO 4 2- > Cl - > NO 3 - >F - > PO 4 3- . The water quality (WQ) index (I) model is a widely used instrument for assessing the quality of drinking water. To overcome low precision, and significant errors in the traditional single prediction model, the article proposes an Entropy ( E )-based classification model for WQ prediction. Various additional classes were established from the output map of the designated EWQI region: of which 42.85 % falls under good and 57.14 % belongs to poor category of water. The accuracy assessment was done by GIS, and based on the overall accuracy, the obtained map was 85.46 % accurate to field value. The findings indicated that an immediate and holistic strategy is required to mitigate various anthropogenic stressors in the study area. The MCDM approach namely, Technique of Order of Preference by Similarity to Ideal Solution (TOPSIS) was utilized to identify the best site. Based on this watershed prioritization, it can be said that site N-(1), (2), (7), and (3) are the polluted ones. Global climate change and rise of human population lead to extensive land-use changes, expanding urbanisation and industrial activity, have already heightened the risk of pollution at these 4 places. Moving forward, Synthetic Minority Over-Sampling Technique (SMOTE) and the application of Support Vector Machine (SVM) investigation has been put into practice, which not only highlights the promise of data-driven approaches in hydrology but also provides insightful information for managing water resources. The analysis of the model findings, spanned a value between 79 and 391, indicating good to unsuitable drinking WQ. However, results revealed that out of the total samples, 2 locations are within the good class, 2 under poor/ very poor and remaining, 7 locations can be utilized under unfit for drinking category. It is noticed that the degradation of WQ at 7 sites, is exacerbated by the disposal of domestic and industrial solid waste, including plastics, polythene bags, paper waste and food waste. The findings highlights the value of Machine Learning (ML) approaches in improving water quality forecast accuracy. Hence, the model’s accuracy was assessed using support vector machine (SVM), by three widely used criteria: root mean square error (RMSE), coefficient of determination (R 2 ) and MAPE. The SVR model typically outperforms, with values ranging from (RMSE) training = 0.08–2.75, (RMSE) testing = 0.04–5.79; (MAPE) training = 0.01–0.59, (MAPE) testing = 0.03–1.2; and R 2 = 0.91–1.0. This impressive accuracy highlights their superiority over rival models in both training and testing, demonstrating their efficacy in WQ predicting. The present research recommends combining Entropy, TOPSIS, SMOTE and SVM modelling methodologies in drinking forecasting with surface water appropriateness since they produce favourable and consistent results. The outcomes of this study will hold significant worth for upcoming researchers as it will reduce the duration and expenses of analysis through the utilization of these algorithms in predicting the sustainability of surface water potability. Therefore, the key findings have significant consequences for WQ processes, and notably, provides valuable insights for ML-based water quality assessment, aiding researchers and professionals in decision-making and management. • This paper gives a brief literature study, analysis, and comparison of the research done in river water quality evaluation using Entropy, MCDM and ML models and techniques. • Valuable contribution to the field of water quality prediction in Brahmani River Basin, Odisha. • Monitoring of the spatial and temporal evolution of physicochemical parameters. • Proposed ML classification model (SVM) for water quality prediction with 15 physicochemical parameters and also tested the algorithm by employing hyperparameter optimization. • Applied Entropy for data normalization, TOPSIS for feature and assigning selection, and SMOTE to address and evaluate water quality deterioration.