Linear and non-linear modeling for the investigation of gas chromatography retention indices of alkylbenzenes on Cit.A-4, SE-30 and Carbowax 20M

Models for theoretical prediction of gas chromatographic retention indices of alkylbenzenes on different polar stationary phases at different temperatures were established by linear and non-linear methods. The linear model is multivariable linear regression. The non-linear models include artificial neural networks (ANNs) and multivariable polynomial regression. The performances of different models were compared. The result of ANN is superior over other methods. The 170 data belong to 46 alkylbenzenes at different temperatures on three different stationary phases. In order to describe the structure of alkylbenzenes, we proposed a new method to describe them. Each alkylbenzene was uniquely represented by a simple set of six numeric codes. Three different stationary phases are Cit.A-4, SE-30 and Carbowax 20M, their polarity were characterized by McReynoldsʹ constant. The eight input parameters included a set of six numeric codes, McReynoldsʹ constant of the different stationary phases, and temperature. The output parameter is retention index. The neural networksʹ architecture and the learning times were optimized. The optimum ANN could give excellent prediction results.