Application of Machine Learning in Predicting Heavy Metal Uptake by Activated Carbon Adsorbents

Abstract

The contamination of water resources by transition metals such as manganese (Mn²⁺), cobalt (Co²⁺), and copper (Cu²⁺) poses significant environmental and health concerns, necessitating the development of sustainable treatment solutions. This study explores the use of activated carbon derived from reed biomass as a low-cost, eco-friendly adsorbent for metal removal. An Artificial Neural Network (ANN) model was developed using a dataset of 435 experimental entries and trained on seven input variables: solution pH, contact time, initial ion concentration, adsorbent dosage, specific surface area (SSA), point of zero charge (pHpzc), and surface functional group intensity (SFG). The ANN, optimized using the Levenberg–Marquardt algorithm with one hidden layer of eight neurons, demonstrated high predictive accuracy, achieving R² values of 0.949 (Mn²⁺), 0.948 (Co²⁺), and 0.923 (Cu²⁺). Sensitivity analysis indicated that pH, contact time, SSA, and SFG were the most influential factors. A user-friendly graphical interface was also developed for real-time adsorption predictions. These findings highlight the effectiveness of reed-derived activated carbon and the ANN model as robust tools for forecasting and optimizing heavy metal removal from wastewater