Akademik Veri Yönetim Sistemi
International Journal of Environmental Science and Technology, cilt.22, sa.16, ss.16837-16864, 2025 (SCI-Expanded, Scopus)
This work aims at investigating the usability of classical regression analysis, Backpropagation- artificial neural network, and artificial neural network training with metaheuristic algorithms models for prediction of Hg(II) removal from aqueous solutions by drinking water treatment sludge. The classical regression analysis approach with seven different regression functions (Quadratic, Power, Exponential, Inverse, Ln, Linear, and S) was applied for the prediction of Hg(II) adsorption and quadratic function with Nash–Sutcliffe value of 0.951 for all data set significantly outperformed the others. The Adaptive Guided Differential Evolution, Escaping Bird Search, Grey Wolf Optimizer, JAYA, Marine Predators Algorithm, Phasor Particle Swarm Optimization, Stochastic Fractal Search, Success-History Based Parameter Adaptation for Differential Evolution, Teaching Learning Based Artificial Bee Colony, Transit Search algorithms were applied to improve the prediction performance of the artificial neural network model, called hybrid artificial neural network model. The Nash–Sutcliffe values of the optimal hybrid artificial neural network model, derived from independent simulations, were 0.992 and 0.997 for the test data and all data, respectively. Furthermore, the fitness of all models to the experimental data was evaluated, and it was determined that the hybrid artificial neural network model exhibited superior predictive performance compared to the classical regression analysis-quadratic function and Backpropagation- artificial neural network models. The research findings highlight the predictive capability of machine learning models that capture the nonlinear relationships in the parameters and the novel methods for predicting Hg(II) adsorption behavior that replace laborious and time-consuming experimental tests.