Abstract
In this study, we utilized a comprehensive dataset comprising over 19,000 data points with inputs represented by coordinates (x, y) and the corresponding output denoted as concentration (C). The case study was analysis of mesoporous material for adsorption separation of target solute from aqueous solution. Mass transfer and machine learning evaluations were carried out to obtain separation efficiency. Our objective was to develop predictive models for C using three distinct base models: decision tree (DT), support vector regression (SVR), and Gaussian process regression (GPR). To enhance the predictive performance of these base models, we employed the ensemble method AdaBoost, which combines their outputs to yield more accurate predictions. Hyper-parameter optimization was achieved through particle swarm optimization, allowing us to fine-tune the models for optimal results. The results of our experiments demonstrate promising performance across the ensemble models. Specifically, AdaBoost combined with decision tree (ADA-DT) yielded an impressive R(2) of 0.96984 and a mean squared error (MSE) of 7.9407E+00. AdaBoost combined with SVR (ADA-SVR) achieved an even higher R(2) score of 0.97148 with an MSE of 7.6362E+00, while AdaBoost combined with GPR (ADA-GPR) produced a commendable R(2) score of 0.95963 with an MSE of 8.16352E+00.