Machine learning method for simulation of adsorption separation: Comparisons of model's performance in predicting equilibrium concentrations

Abstract

In this work, we implemented different models for predicting adsorption separation of a dye from aqueous solution using porous materials. The equilibrium data of solute concentrations were collected from resources and used in the models for training and verification purposes to develop the models. For prediction of the equilibrium solute concentrations (Ce), we used tree models: Multi-layer Perceptron (MLP), Passive aggressive regression, and Decision Tree (DT) Regressor. In the modeling, we considered the adsorbent dosage as well as solution pH as the input parameters to the model, and the model was able to generate the output values, i.e., equilibrium concentrations based on the input variables. The evaluation of the models’ performances revelated that the final R2 scores are 0.99, 0.98, 0.93 for DT, MLP and Passive-Aggressive, respectively and a very low RMSE of 0.055 for decision tree that shows this model is the best among models used in this study. Indeed, decision tree model is recommended among the other three models to be employed for correlation of adsorption equilibrium data.