Table 3.

ANN modeling instructions for the present study

Model parameter	Instruction/activity report
Input variables	The predictor variables are pH, EC, TDS, TSS, TH, Na, K, Ca, Mg, Cl, SO4, HCO3, NO3, Fe, Zn, Ni, Cr, and Pb. However, for the predictions of NO3, Fe, Zn, Ni, Cr and Pb, the predicted is made the dependent variable and all others used as predictors. Meanwhile, for the prediction of WPI and PIG, all predictor variables were utilized
Output variable	The predicted parameters are NO3, Fe, Zn, Ni, Cr, Pb, WPI, and PIG
Hidden layer activation function	Hyperbolic tangent
ANN type	Multilayer perceptron (MLP)
Rescaling of covariates	Normalized
Partitioning of dataset	Randomly assigned cases based on relative number of cases: NO3: Training (75%), Testing (25%), and Validity of cases (100%) Fe: Training (70%), Testing (30%), and Validity of cases (100%) Zn: Training (90%), Testing (10%), and Validity of cases (100%) Ni: Training (80%), Testing (20%), and Validity of cases (100%) Cr: Training (85%), Testing (15%), and Validity of cases (100%) Pb: Training (85%), Testing (15%), and Validity of cases (100%) WPI: Training (80%), Testing (20%), and Validity of cases (100%) PIG: Training (80%), Testing (20%), and Validity of cases (100%)
Number of hidden layers	One (1)
Output layer activation function	Hyperbolic tangent
Number of units	Automatically computed
Rescaling of scale-dependent variables	Adjusted Normalized (Correction = 0.02)
Type of training	Batch
Optimization algorithm	Scaled conjugate gradient (SCG)