Data-driven soft computing modeling of groundwater quality parameters in southeast Nigeria: comparing the performances of different algorithms

In recent decades, the simulation and modeling of water quality parameters have been useful for monitoring and assessment of the quality of water resources. Moreover, the use of multiple modeling techniques, rather than a standalone model, tends to provide more robust and reliable insights. In this present paper, several soft computing techniques were integrated and compared for the modeling of groundwater quality parameters (pH, electrical conductivity (EC), total dissolved solids (TDS), total hardness (TH), modified heavy metal index (MHMI), pollution load index (PLI), and synthetic pollution index (SPI)) in Ojoto area, SE Nigeria. Standard methods were employed in the physicochemical analysis of the groundwater resources. It was found that anthropogenic and non-anthropogenic activities influenced the concentrations of the water quality parameters. The PLI, MHMI, and SPI revealed that about 20–25% of the groundwater samples are unsuitable for drinking. Simple linear regression indicated that strong agreements exist between the results of the water quality indices. Principal component and Varimax-rotated factor analyses showed that Pb, Ni, and Zn influenced the judgment of the water quality indices most. Q-mode hierarchical and K-means clustering algorithms grouped the water samples based on their pH, EC, TDS, TH, MHMI, PLI, and SPI values. Multiple linear regression (MLR) and artificial neural network (ANN) algorithms were used for the simulation and prediction of the pH, EC, TDS, TH, PLI, MHMI, and SPI. The MLR performed better than the ANN model in predicting EC, TH, and TDS. Nevertheless, the ANN model predicted the pH better than the MLR model. Meanwhile, both MLR and ANN performed equally in the prediction of PLI, MHMI, and SPI.