Artificial Neural Network Model for Predicting Iron Concentration in Groundwater Using Water Quality Parameters

Abstract

Elevated concentrations of dissolved iron (Fe²?) in groundwater represent a widespread issue that can degrade water quality, cause aesthetic problems, and adversely affect water supply systems and infrastructure. Iron behavior in groundwater is governed by complex and nonlinear interactions among physicochemical parameters, including pH, temperature, total dissolved solids (TDS), turbidity, and dissolved oxygen (DO), which limits the effectiveness of simple linear predictive approaches. In addition, the availability of field data covering a sufficiently wide range of iron concentrations is often restricted, posing challenges for developing robust predictive models. To address these limitations, this study employed laboratory-prepared Synthetic Groundwater (SGW) designed to replicate the chemical characteristics of natural groundwater under controlled conditions. A total of 37 SGW samples were generated, representing dissolved iron concentrations both below and above the national groundwater quality standard. Key water quality parameters were systematically measured and compiled into a modeling dataset. Correlation analysis confirmed that the selected input parameters exerted a significant influence on dissolved iron concentration, supporting the reliability of the dataset and the modeling framework. An Artificial Neural Network (ANN) model was subsequently developed, trained, and validated using a cloud-based computational environment. Model evaluation demonstrated strong predictive performance, as indicated by low Mean Squared Error (MSE) values and high coefficients of determination (R²), reflecting high accuracy and generalization capability. To enhance practical applicability, the trained ANN model was integrated into a web-based prediction application capable of providing real-time estimates of iron concentration and classifying groundwater quality as “Safe” or “Not Safe” according to national regulatory standards. The results indicate that the integration of SGW-based data generation with ANN modeling constitutes a reliable and practical approach for groundwater iron prediction and quality assessment.