Prediction of Dye Removal Using Machine Learning Techniques

Abstract

This study aims to predict the removal efficiency of methylene blue dye using experimental data collected from adsorption processes involving acorn-based biosorbents. A comparative evaluation of four machine learning algorithms (Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), Random Forest, and XGBoost) was conducted to determine the most suitable modeling approach. Two ANN architectures, with single and dual hidden layers respectively, achieved the highest predictive accuracy, with R2 values of 0.93 and 0.87. While XGBoost demonstrated better performance (R2 = 0.64) than Random Forest (R2 = 0.61), both ensemble models provided moderately accurate predictions. In contrast, the LSTM model performed poorly (R2 = 0.44), likely due to the non-sequential structure of the dataset. These findings underscore the potential of ANN-based models for accurately capturing nonlinear relationships in adsorption systems and also demonstrate the viability of alternative ensemble learning methods for predictive environmental modeling.