Prediction of programmed-temperature retention indices of polycyclic aromatic hydrocarbons in the Lee index scale by artificial neural network

Abstract

A feed-forward artificial neural network (ANN) model was used to predict the programmed-temperature retention indices RIs in the Lee index scale of polycyclic aromatic hydrocarbons (PAHs). The data used in this paper include 96 RIs in Lee index scale of 48 parent and alkylated PAHs obtained on SE-52 and DB-5 slightly polar stationary phases with three different temperature programmes. Four parameters: boiling point, molecular weight, connectivity index and F-number were used as input parameters. The data containing 96 RIs were randomly divided into three sets: a training set (including 32 RIs obtained on SE-52 stationary phase), a validation set (including 16 RIs for the same SE-52 phase) and testing sets (including 48 RIs obtained on DB-5 stationary phase and with two different temperature programme, 30 at one and 18 RIs at the other programmes, respectively). The structures of networks and the number of learning epochs were optimized. The best network structure is 4-6-1. The optimum number of learning epoch is 1000. The results obtained in this study showed that the average percentage deviation between the predicted ANN values and the experimental values of Lee retention indices for. the validation and two testing sets were 1.42% on the SE-52 and 1.32 and 1.43% on the DB-5 stationary phases, respectively. The result illustrated that the prediction performance of ANN in the field of investigating the programmed-temperature retention behavior of polycyclic aromatic hydrocarbons is very satisfactory.