An integrated water quality assessment and its prediction utilizing the CWQI, CRITIC based-improved CWQI, sensitivity analysis, multivariate techniques, machine learning algorithms and GIS approaches to determine pollution intensities in the river basin (Odisha).

The Mahanadi River is a vital water resource for the Odisha region, supporting both local agriculture and drinking water needs. However, increasing human activities, including religious practices and inadequate waste management, pose significant challenges, leading to bacterial contamination and deteriorating water quality, which demands urgent attention. A hydrogeochemical study was conducted to identify the processes that control surface water chemistry and examine the water sources' quality for domestic and agricultural purposes. Thirteen water quality samples were taken throughout the monsoon season from 11 monitoring sites in the research region. The study utilized for a period between 2017 and 2024. The samples' primary ions and physicochemical properties were examined. The surface waters had a pH as alkaline (7.74-7.89), while turbidity exceeds 5 NTU for all tested sites. 100% of the rivers have TDS and coliform concentration, that records above the WHO guidelines, indicating maximum mineralization, hence saline water. Calcium (Ca2+) is the dominant cation composition while, Bicarbonate (HCO3-) remained the primary anion in the detected season. The Water Quality Index (WQI) is a distinctive and successful rating system for evaluating water quality. The study combines the comprehensive (C) WQI method and the improved CWQI method based on CRITIC-(C)R to evaluate the water quality. Based on the CWQI, the water quality in the study area ranged from excellent to very poor for domestic use. Conversely, the 63.64% suggests poor water quality that may have led to increased sedimentation and nutrient runoff, potentially worsening bacterial contamination. Water pollution at 7 stations, is mostly attributed to the discharge of untreated industrial and urban effluents directly into rivers, without undergoing any form of water treatment. The improved CWQI findings of the study indicate that the water from the river is deemed appropriate for drinking purposes, as determined at 27.27%. However, 72.73% of tested locations are deemed entirely unsuitable. Moreover, the spread of modern agriculture with the extensive application of chemical fertilizers significantly lowers the river water health. Sensitivity analysis shows that turbidity, TDS, TH, and coliform are more sensitive compared to other indicators, indicating that they are the main elements affecting the assessment's findings. To evaluate the variances in surface water quality, this study uses multivariate statistical methods (MSTs), such as principal component analysis (PCA) and discriminant analysis (DA). Using DA, five water quality metrics such as coliform, turbidity, TDS, TH, pH, HCO3-, Cl-, and SO42- were found during the spatial evaluation, and their assignment rate was 100%. A total of four principal components (PCs) explained 87.60% of the observed variance. The effectiveness of PCA in evaluating and analyzing large, intricate data sets on water quality and identifying the causes and contributing variables of water pollution, the decline in surface water quality was shown to be largely caused by surface runoff, storm runoff, soil weathering, leaching, residential discharge, and agricultural runoff. To enhance the precision of water quality assessment, the study employed different established machine learning (ML) techniques, using the model's performance based on factors such as uncertainty, sensitivity, and reliability. Nearly eight sampling locations had poor or extremely poor water quality, according to GAPSO-WQI results. During the monsoon season, the water quality deteriorated the most at places like N-(2) and (5-11). Additionally, water quality ranges from excellent to inappropriate, according to GMDH-WQI measurements. According to the study's findings, N-(7) and (11) are the most contaminated among the 13 sampling locations. It is observed from different approaches that many water quality parameters including turbidity, TDS, Ca2+, sulphate, TH, and coliform are found to scored higher and exceeded the permissible limits prescribed by WHO at the polluted locations. In regards to Firefly algorithm, the obtained value exhibited 36.36%, that rates bad water quality. Here, water quality assessment revealed that surface water quality ranged from excellent to very bad, indicating that eight sites cross the desired threshold (< 50). The Adaptive Neuro-Fuzzy Inference System (ANFIS) model was developed in MATLAB R2021a using Gaussian membership functions and a hybrid learning algorithm (backpropagation + least squares). Model performance was evaluated using statistical metrics, namely, mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The proposed ANFIS model achieved high predictive accuracy, with MAPE = 9.09%, RMSE = 17.08%, and R2 = 0.8756 during testing, and further improvement during validation (R2 = 0.9927). The study demonstrates that ANFIS provides a robust and accurate approach for forecasting electrical conductivity and understanding the nonlinear interactions among water quality parameters. Finally, this new study offers a methodical assessment and highlights the applicability and effectiveness of various approaches for achieving sustainable management of water resources while identifying the causes of changes in water quality. As a result, it exhibits a noteworthy contribution to eco-balance and water management.