Abstract
A set of Kováts retention indices (
I
) of 73 petroleum hydrocarbons has been analyzed statistically. The retention indices were related to two molecular encoding descriptors generated by Dragon software, total path count (TPC) and sum of atomic polarizabilities (
S
P
), which are simulated for non-polar and polar dimensions. Principal-components analysis (PCA) was used to improve the computation of the descriptors and to achieve descriptor orthogonality, to obtain more obvious differences between the homologous series examined. Two and three-dimensional models were constructed by use of Unscrambler 9.7 software, using the two effective descriptors. PCA was used to improve the accuracy of both fitting and prediction of the quantitative structure–retention (QSRR) models constructed, with predictive
R
values of 0.992 and 0.996 for
I
. The accuracy of the constructed models was tested by prediction of unknown
I
values for some hydrocarbons.