Key points are not available for this paper at this time.
Abstract Accurate prediction of water quality parameters is important for management of freshwater resources and for ensuring sustainable water. Electrical conductivity (EC) is an important parameter that indicates ionic concentration, salinity, and potential contamination in both groundwater and surface water. However, conventional monitoring approaches are hindered by high cost, time consumption, and limited spatial coverage, especially in developing countries like Uganda. This study designed and validated an Adaptive Neuro-Fuzzy Inference System (ANFIS) model to predict EC from physicochemical and heavy metal parameters of water sources in Kampala and Mbarara Districts, Uganda. Sixty water samples with physicochemical parameters and heavy metal concentrations from groundwater and surface water were used. The data were randomly divided into training (80%), testing (10%), and validation (10%) subsets. The ANFIS model was built in MATLAB R2021a using Gaussian membership functions and a hybrid learning algorithm (backpropagation and least squares estimation). Model prediction was assessed using Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the Coefficient of Determination (R²). The ANFIS model achieved high predictive accuracy, with RMSE = 91.53 ± 3.74 µS/cm, MAPE = 6.34 ± 0.34%, as well as the highest R² = 0.943 ± 0.009 (after 10-fold cross-validation), when compared with Artificial Neural Network (R² = 0.884 ± 0.010) and Gene Expression Programming (R² = 0.840 ± 0.009). Statistical analysis revealed that the Total Dissolved Solids (TDS) and Total Suspended Solids (TSS) are the most significant predictors of EC (p < 0.001). Additionally, sensitivity analysis showed a strong positive correlation between TDS and EC and a negative correlation with TSS, while heavy metals exhibited a weak direct influence. Findings from this study reveal that ANFIS can be used as an effective framework for EC prediction and early contamination detection in resource-limited areas.
Sanusi et al. (Tue,) studied this question.