Surface water quality assessment for drinking and pollution source characterization: A water quality index, GIS approach, and performance evaluation utilizing machine learning analysis

The Mahanadi River Basin, a vital and largest river of Odisha, faces increasing surface water quality deterioration due to anthropogenic activities. The sustainable management of surface water is made more difficult by the dearth of thorough research on its chemical composition and the geochemical mechanisms affecting its hydrochemistry. The research aims to fill the gap by considering a total of 13 surface water samples, which were collected and analyzed using standard methods. Evaluating the effectiveness and dependability of classification algorithms in identifying changes in water quality is crucial since accurate information is required to improve decision-making. This study sought to evaluate the region's surface water quality and sources of contamination using machine learning (ML) methods such as Logistic Regression (LOR), Random Forest (RF), Artificial Neural Network (ANN), Support Vector Machine (SVM), Decision Tree (DT), and K-Nearest Neighbor (KNN). Two indicators, including turbidity and coliform, exceeded the WHO guidelines, according to the samples' hydrochemistry study. The study finds distinct trends in the concentrations of cations and anion. The mean concentration of cations is rated as Mg 2+ > Ca 2+ > Na + > K + , while, the major anions are typically showing a ranking of HCO 3 - > Cl - > SO 4 2- > NO 3 - . The geochemical processes influencing the chemistry of surface waters are primarily driven by the weathering of silicate minerals. An increase in mineralization, is particularly important for magnesium silicates and carbonates. Various water quality indices (WQIs), implementing machine learning’s models including ANN, SVM, DT, RF, LOR), and KNN, jointly revealed that 36.36% of samples reflects excellent to good water quality, while 63.64% renders water quality status as unsuitable, which were mostly found in the central, south-eastern, and northeastern areas. Subsequently, geogenic and anthropogenic influences have a considerable impact on the surface water chemistry at seven water locations, according to the various ML analyses. According to the results, turbidity, TDS, TH, and coliform are important factors that shape different clusters of water quality. The random forest (RF) model with the highest accuracy and superior performance, that pertains to a score of R 2 as 0.986, was being referred as best prediction model, while logistic regression (LOR) corresponds to R 2 = 0.98, KNN as (R 2 = 0.968), ANN as (R 2 = 0.955), and finally, SVM includes R 2 = 0.9 2 8. Basically, all approaches cumulatively predicted surface WQI with lower accuracy. Overall, the LOR, KNN, ANN, and SVM algorithms were all surpassed by the RF algorithm. The proposed approaches are robust in characterizing diffusive pollution loads. Turbidity and Coliform loadings are estimated to increase in the relevant area. Deforestation, human-induced factors, and intensive agricultural activities are major contributors. Machine learning techniques such as RF, KNN, ANN, SVM, LOR, and DT, have a great potential for estimating significant pollution loads. RF algorithm exhibited superior performance, and outperformed than the other models. The study sets the stage for future investigations.