Machine learning-assisted prediction of engineered carbon systems' capacity to treat textile dyeing wastewater via adsorption technology.

Abstract

Dyes are widely used in industries like printing, cosmetics, paper, leather processing, textiles, and manufacturing to add color to products. However, improper disposal of dyes into wastewater has raised major concerns due to their harmful effects on plants, animals, and humans. Using engineered carbon systems (ECSs) to treat dye-contaminated wastewater has shown promise for sustainable waste management. Dye adsorption on ECSs is a complex, non-linear process, making it essential to understand ECSs' dye removal capabilities through a modeling framework that includes experimental and environmental factors. To support this, a database of ECSs used in dye removal from textile wastewater was compiled. Twelve machine learning models, including XGBoost, Light Gradient Boost, Random Forest, Gradient Boost, CatBoost, AdaBoost, Decision Tree, Artificial Neural Network, K-Nearest Neighbor, Support Vector Machine, Huber, and Ridge Regressor, were applied to analyze ECSs' dye removal potential. Out of all the models, XGBoost exhibited the highest coefficient of determination (R²) of 0.986 during the training and 0.978 during testing, alongside the lowest prediction error (MSE) of 0.01 and 0.136 in the training phase and testing phase. The quantity of ECS, concentration of dye (C_o), and pH of wastewater highly influenced the adsorption process. The optimization results indicated the highest affinity of direct, reactive, and dispersed dyes towards ECSs in the acidic solution. In contrast, the maximum adsorption of Basic and VAT dye on ECSs was found in the alkaline solution. The partial dependence analysis provided valuable insights into the interaction between ECS dose and water matrix parameters that can lead to efficient extraction of dyes from aqueous matrices.