Application of MLR-PRN model for estimation of arsenic concentration in drinking water: a case study for Izmir City

Arsenic in drinking water can have serious adverse health effects depending on consumption. This study includes the comparison of two neural networks, Pattern Recognition Network (PRN) and Cascade Forward Neural Network (CFNN), that uses chloride, pH, and electrical conductivity variables at their inputs for estimation of arsenic in drinking water at the water distribution stations in Izmir. The data set consists of arsenic, electrical conductivity, iron, aluminum, pH, and chloride chemical variables, measured every two weeks (average) and taken from 16 water distribution points. Multiple linear regression (MLR) analysis showed that electrical conductivity is the most significant variable for arsenic estimation. The MAE (Mean Absolute Error), NSE (Nash-Sutcliffe Efficiency), IA (Index of Agreement), and R-2 (coefficient of determination) values obtained with the PRN model were 0.653, 0.789, 0.944, and 0.802, respectively. These results showed that arsenic concentration in drinking water could be estimated using chloride, pH, and electrical conductivity.