Application of Artificial Neural Network Models for Predicting Water Quality Index

This study discusses the development and validation of an Artificial NeuralNetwork (ANN) model in estimating water quality index (WQI) in the Langat RiverBasin, Malaysia. The ANN model has been developed and tested using data from 30monitoring stations. The modeling data was divided into two sets. For the first set, ANNswere trained, tested and validated using six independent water quality variables as inputparameters. Consequently, Multiple Linear Regression (MLR) was applied to eliminateindependent variables that exhibit the lowest contribution in variance. Independentvariables that accounted for approximately 71% of the variance in WQI are DissolvedOxygen (DO), Biochemical Oxygen Demand (BOD), Suspended Solids (SS) andAmmoniacal-Nitrate (AN). The Chemical Oxygen Demand (COD) and pH contributedonly 8% and 2% to the variance, respectively. Thus, in the second data set, only fourindependent variables were used to train, test and validate the ANNs. We found that thecorrelation coefficient given by six independent variables (0.92) is only slightly better inestimating WQI compared to four independent variables (0.91) which demonstrates thatANN is capable of estimating WQI with acceptable accuracy when it is trained byeliminating COD and pH as independent variables.