Identification of fatty acids from açai in mistures subjected to crossflow microfiltration using Artificial Neural Networks

Abstract

This article proposes the use of Artificial Neural Networks as an alternative tool to classify mixtures of fatty acids from açai subjected to crossflow microfiltration. The neural networks were trained with the Levenberg-Marquardt and Bayesian Regularization supervised learning algorithms, using the software MATLAB and datasets from literature, related to pressure, flow speed, time and transmembrane flow rate. The evaluation criteria were consists on analyzing the confusion matriz, the error histogram and the performance graphics. The neural network that used the Bayesian Regularization had 99,1\% accuracy for the mixtures with water/oleic acid and water/palmitic acid and also 96,5\% accuracy when a third class was added to the datasets, containing water and both the fatty acids, showing the validity of the tool.