A QSPR–ANN-driven predictive framework for assessing organic micropollutant rejection in forward osmosis systems

Forward Osmosis (FO) is the subject of many current studies, given existing and future conditions around the world. This work is the continuation of the series of research that implicates Artificial Neural Networks in the processes of membrane separation. Three databases (with the same size of 193 points), two learning algorithms, two function transfers, five subdivisions of the database, and eleven (11) inputs were used with the aim to extract the optimal QSPR-NN model which is chosen based on the best values of coefficient of correlation (R) and the Root Mean Squared Error (RMSE). QSPR-NN (Quantitative Structure-Property Relationships - Neural Networks) model obtained was characterized by eleven (11) neurons on the input layer, fourteen (14) neurons in the hidden layer, and one (1) neuron in the output layer, the Bayesian regularization (Trainbr) was the learning algorithm, tangent sigmoid (Tansig), and purelin were the transfers functions for the hidden and output layers respectively. The performance of the QSPR-NN optimal model obtained was demonstrated with a higher value of (R = 0.9895) and low Root Mean Squared Error (RMSE = 4.3683%), and other errors as RER and RPD more than 2.5 and equal to 21.4356 and 3.4290 respectively, the (NSE) more than 0.9. Furthermore, the comparison with other work in the same orientation demonstrated the excellence of our model developed in this work compared to the others.