Modeling Surface Water Quality Using the Adaptive Neuro-Fuzzy Inference System Aided by Input Optimization

Modeling surface water quality using soft computing techniques is essential for the effective management of scarce resources and environmental protection. The development accurate predictive models with significant input parameters inconsistent datasets still a challenge. Therefore, further research needed to improve performance models. This study presents methodology dataset pre-processing optimization reducing modeling complexity. objective this was achieved by employing two-sided detection approach outlier removal an exhaustive search method selecting inputs. Thereafter, adaptive neuro-fuzzy inference system (ANFIS) applied electrical conductivity (EC) total dissolved solids (TDS) in upper Indus River. A larger 30-year historical period, measured monthly, utilized process. prediction capacity developed estimated statistical assessment indicators. Moreover, 10-fold cross-validation carried out address overfitting issue. results indicate that Ca2+, Na+, Cl− are most relevant inputs be used EC. Meanwhile, Mg2+, HCO3−, SO42− were selected model TDS levels. optimum ANFIS EC data showed R values 0.91 0.92, root mean squared error (RMSE) 30.6 µS/cm 16.7 ppm, respectively. structure comprises hybrid training algorithm 27 fuzzy rules triangular membership functions Gaussian curve modeling, Evidently, outcome present reveals aided optimization, suitable technique simulating water. It could minimizing complexity elaborating proper mitigation measures.