Group contributions methods have been successfully applied over the years to accurately predict primary properties such as Tc, Pc, and boiling point for compounds with a known structure but no experimental data. Past approaches have relied on regression of a large set of data collected over wide conditions and for a large number of chemicals from many chemical families. This approach was influenced by the availability of computing hardware and software techniques. A more accurate approach would be to customize the regression by selecting a subset of data appropriate to the compound with respect to its size, shape, and functional groups.

Using the known chemical structure, the database is scanned to identify all compounds which are part of the same chemical family (i.e.., having approximately the same type and number of functional groups). This set of data is further analyzed to identify a set of frequently occurring functional groups. Using this set of data and the published group contribution factors, a set of new group contribution factors are generated for these dominant functional groups. Functional groups occurring less frequently remain unchanged. This approach has a high probability of producing a more accurate estimate of the primary physical properties as compared to the original group contribution factors which were regressed from the complete data set.

This technique has been used to refine the parameters used in Constantaniou-Gani's group contribution method as well as Wilson-Jasperson's atomic contribution method. An analysis of the improvement in the prediction of the critical temperature and pressure using these methods and the dynamic refitting of the factors will be presented.