Separation of organic compound from water using membrane process: Hybrid machine learning-based modeling and validation

Abstract

Modeling was performed for separation of organic compounds from aqueous solution using porous membranes. Ensemble methods based on decision trees have gained significant attention in the field of predictive modeling due to their capability to capture complex relationships and handle high-dimensional data. In this study, we explore the effectiveness of three popular ensemble methods, namely Random Forest (RF), Extra Trees (ET), and Gradient Boosting (GB), for predicting the concentration of a chemical compound (C) in membrane system, using the radial distance (r) and axial distance (z) as input features. To optimize the performance of the ensemble models, we employ a novel hyper-parameter meta-heuristic technique called Political Optimizer (PO). Furthermore, to ensure the robustness of the models, we incorporate an outlier detection mechanism based on the Z-Score. The experimental results demonstrate the efficacy of the ensemble methods in predicting the concentration of the chemical compound. The PO-ET model achieves an impressive coefficient of determination (R2) score of 0.999, with a root mean squared error (RMSE) of 0.021 and a mean absolute percentage error (MAPE) of 6.60e+11. Similarly, the PO-GB and PO-RF models exhibit strong predictive performance, with R2 scores of 0.998 and 0.997, respectively, and corresponding RMSE and MAPE values.