QSPR Analysis of Boiling Point of Chemical Compounds
Barbara Dębska · 2002
A selected data mining methods have been developed to model the relationships between the structure of organic compounds and their properties. Molecular graph descriptors represent valuable structural descriptors that can be used with success in developing QSPR model. In this study we have used: four valance molecular connectivity indices (1χv, 2χv,3χv, 4χv), a second order Kappa shape index (2 K), molecular weight and dipole moment. The database included seven structural descriptors and experimental value — boiling point for each compound. The paper presents the proposed cluster analysis and neural network methods used to estimate the boiling points of chemical compounds. Back-propagation 7–4–1 neural network architecture predicted boiling points of aliphatic hydrocarbons with average absolute errors of 1,55 [K] – 4,85 [K], respectively.