ANN modeling of extraction kinetics of essential oil from tarragon using ultrasound pre-treatment

In this paper, an artificial neural network (ANN) modeling is utilized to predict kinetics of essential oils extraction from tarragon (Artemisia dracunculus L.) using Ultrasound pre-treatment with Clevenger. A three-layer perceptron artificial neural network was created to predict the extract model with an error back-propagation algorithm. To design the ANN model, ultrasound power, sonication time, extraction time and their interactions were considered as input vectors while the extraction yield of essential oils was considered the model output. The performance of the network was optimized by varying the number of nodes in the hidden layer to achieve the best ANN architecture for output prediction. The performance of different ANN architectures was obtained as error (mean squared errors: MSE) and goodness of fit (determination coefficient: R2) parameters. The results showed that the best prediction performance belonged to 3-7-1 ANN architecture (0.0008 normalized MSE and 0.99 R2) which means that it is possible to predict the extraction yield of essential oils with an acceptable error having the three input parameters. The main extracted compounds by two methods at different conditions were estragole (76.6-83.0%), (Z)-β-ocimen (5.7-8.7%), (E)-β-ocimen (5.2-7.9%).