New generation neurocomputing learning coupled with a hybrid neuro-fuzzy model for quantifying water quality index variable: A case study from Saudi Arabia

Ensuring availability in terms of quality and quantity and sustainable management of safe, affordable drinking water is one of the integral parts of envisioning the 2030 Sustainable Development Goals (SDGs). Saudi Arabia faces many challenges in terms of water supply, inadequate water resources, and distribution due to low rainfall throughout the year. An uncertain water quality index (WQI) has been quantified to monitor water resource quality and management. The current study developed six different computational models WQI, namely: Generalized regression neural network (GRNN), Elman Neural Network (Elm NN, considered as a new generation learning tool), Feed Forward Neural Network (FFNN), Support Vector Machine (SVM), Linear Regression (LR), and Neuro-Fuzzy (NF). The experimental data were collected from 40 sampling locations. The obtained physicochemical variables (pH, EC, Turb, TDS, COD, Cl, NH 3 , PO4 − , N/NO3 − , SO4 − , and TPC) were subjected to feature sensitivity technique, and the model combinations were determined based on sensitivity analysis and principal component analysis (PCA). Goodness-of-fit, error criteria, and mean bias coupled with visualization methods were used to assess the accuracy of the models. The quantified results showed that the NF model surpassed the other models and provided the highest accuracy. NF produced the highest R 2 value of 0.9989 and lowest MAD = 0.0590, MAPE = 13%, and BIAS = −0.0003. The outcomes indicate that the water quality at a few locations requires minor treatment. The techniques employed validated the application of computing intelligence for optimum decision-making. • Hydro-environmental modelling is crucial in Saudi Arabia. New generation learning was used for modelling the WQI variable. • Subsequently, different data intelligent methods were used for comparison. • Sensitivity analysis and PCA was employed for feature extraction. • Results served as a robust multi-decision technique by policymakers.