Predicting solubility in supercritical fluid extraction using a neural network

Abstract

A neural network has been constructed for prediction of the solubility of analytes in supercritical carbon dioxide. Preliminary studies for the input of molecular structure into the network indicates that connectivity indices are adequate to provide structural information in a condensed form. This allows neural networks, which would otherwise be very extensive, to have reduced training times; it also reduces the possibility of memorization of the training data and over-training of the network.