Density prediction of n-alkane and cycloalkane mixtures using calculated molecular descriptors and Levenberg–Marquardt artificial neural network

In this work, we propose a quantitative structure-property relationship (QSPR) approach in order to model the density of different alkane mixtures and also their mixture with cyclohexane over a wide range of temperature and pressures. Levenberg-Marquardt artificial neural network (ANN) was used to link molecular structures and density data. Among a large number of descriptors which were calculated with Dragon software, only 7 significant descriptors were obtained by genetic algorithm based partial least squares (GA-PLS) as the most feasible descriptors, and then they were used as inputs for neural network. These descriptors are: Pressure, Temperature, MW, X2Av, X2A, BEHm3, R7m-A. The data set was randomly divided into three data sets: training (399 point), validation (80 point) and test set (80 point).and the neural network architecture and its parameters were optimized. The prediction ability of the model was evaluated using the validation and test data sets. The mean square errors (MSE) and R2 were 4.4611, 0.9987 for the validation data set and 4.1911, 0.9988 for the test data set, respectively. The obtained results showed the excellent prediction ability of the proposed model in the prediction of density for different alkane and cycloalkane mixtures.