Machine Learning Models for Efficient Adsorption of Congo Red Dye on High-Performance Polyethyleneimine Macroporous Sponge

In this work, machine learning (ML) models were formulated to predict the batch and column adsorption of Congo red (CR) dye on high-performance polyethyleneimine (PEI) based macroporous monolithic sponge (S 100 ) adsorbent, which was designed using the ice-templating technique. Compared to previous PEI-based adsorbents, the fundamental innovation of presently developed sponges is their cost-effective, single-step production technique, which resulted in excellent selectivity and remarkable adsorption ability against anionic contaminants. The dye adsorption percentage for the batch process was predicted as a function of adsorbent dose, contact time, and concentration of dye, whereas the relative concentration for column experiments was forecast as a function of flow rate, concentration of dye, adsorbent dose, and time of contact. The competency of the ML models namely, support vector regression (SVR) and artificial neural network (ANN) was examined against the frequently used conventional multiple regression (MR) model. For batch SVR, ANN, and MR adsorption models, the correlation coefficient (R) values of 0.9955, 0.9972 and 0.8857 were achieved. The R values for continuous SVR, ANN, and MR adsorption models were 0.9991, 0.9994, and 0.9541, whereas the root mean square error (RMSE) values were found as 0.0141, 0.0118, and 0.1034, respectively. The generalized machine learning models best matched the breakthrough experimental data. However, it can be concluded that these ML techniques can help users to estimate how the adsorption system will react to variations in the experimental conditions, giving them a tool to improve the design and operation of adsorption processes, especially employed for wastewater treatment.