Modeling the Retention of Organic Compounds by Nanofiltration and Reverse Osmosis Membranes Using Bootstrap Aggregated Neural Networks

The purpose of the current work was to investigate the use of bootstrap aggregated neural networks (BANN) in modeling the rejection processes of charged and uncharged organic compounds by nanofiltration and reverse osmosis membranes. A total database of 436 rejections of 42 charged and uncharged organic compounds collected from the literature was used to build the BANN model. The training dataset (350 data points) is re-sampled using a bootstrap technique to form different training datasets. For each training set, an INN model is constructed, validated, and tested. The predicted outputs obtained from the developed INN models are then combined together by simple averaging. Good agreement between the predicted and experimental rejections for the BANN model was obtained (the correlation coefficient for the test dataset was 0.9862). The comparison between the BANN, single neural network (SNN), and bootstrap aggregated multiple linear regressions (BAMLR) revealed the superiority of the BANN model (the root mean squared errors for the test dataset were 5.3315 for the BANN, 6.4587 for the SNN, and 18.7834 for the BAMLR).