QSRR PREDICTION OF NITROGEN-CONTAINING POLYCYCLIC AROMATIC COMPOUNDS FROM INCOMPLETE COAL COMBUSTION BASED ON SUPPORT VECTOR MACHINE AND DENSITY FUNCTIONAL THEORY

The quantitative structure-retention relationship (QSRR) models were established for predicting the gas chromatographic retention indices (RI) of nitrogen-containing polycyclic aromatic compounds (N-PACs) by quantum chemistry descriptors, including the total Hartree-Fock energy (E-HF), the eigenvalue of the highest occupied molecular orbital (E-HOMO), the eigenvalue of the lowest unoccupied molecular orbital (E-LUMO), and the dipole moment (mu). The quantum descriptors of N-PACs, selected and calculated by density functional theory (DFT), were used to construct some linear and nonlinear models with the employment of multiple linear regression (MLR), artificial neural network (ANN), and support vector machine based on particle swarm optimization (PSO-SVM). The values of squared correlation coefficients (R-2) between experimental and calculated ones of RI for the test set (12 N-PACs) by MLR, ANN and PSO-SVM were 0.949, 0.805 and 0.982, with max relative error values of 6.72, 13.79 and 4.67%, respectively. The results indicated that the PSO-SVM model was superior to the MLR and ANN models. This investigation also proved that the proposed models can be extended to predict the RI of other unknown N-PACs using quantum chemical descriptors.