Yield Prediction of Supercritical Fluid Extraction of Nigella Sativa using Neutral Networks

Abstract

A feed-forward multi-layer neural network with Levenberg-Marquardt training algorithm was developed to predict yield for supercritical carbon dioxide (SC-CO2) extraction of Nigella sativa essential oil. Yield of extraction depends on these variables: pressure, temperature, and extraction time hence were chosen as the input to the network. Different number of neurons in hidden layer were trained and tested using training and testing data sets. The validating data set was used to determine the network that having lowest mean-squared error (MSE) value and highest regression coefficient. The optimal ANN model, featuring four neurons in hidden layer, demonstrated high predictive accuracy with the lowest MSE of 0.42 ,1.43 and 1.25 for training, validation and test model, respectively. The regression plots indicated high R-values of 0.99641, 0.99513, and 0.98874 for the training, validation, and testing sets, respectively, confirming the model's robustness in predicting experimental data. A very good fitting between the predicted data and experimental data was observed with R2 of 0.9891 indicates ANN shows good accuracy in predicting yield of Nigella sativa.