Neural Network Capability for Retention Modeling in Micellar Liquid Chromatography with Hybrid Eluents

Abstract

Until date no papers have been reported about modeling studies in Micellar Liquid Chromatography by means of neural networks, only classical statistical methods have been used. In this work an overview into the capabilities of neural networks for modeling retention data in Micellar Liquid Chromatography is presented. To solve the problem of capacity factor modeling some parameters have been evaluated: type of activation function, number of neurons in the hidden layer and the use of some input and/or output data transformations. These studies have been carried out with the retention data for twenty-three compounds (benzene derivatives and polycyclic aromatic compounds) in an octyl silica column using micellar mobile phases containing CTAB (hexadecyl-trimethylammonium bromide) as the surfactant and modified with n-propanol. From the results obtained some considerations can be drawn. The selection of the activation function is very important to get good results and the best ones have been obtained when a linear activation function, a recurrent network and a logarithm transformation (logarithm of the capacity factor) have been used.