this work focuses on determining the suitability of functional group properties from the quantum theory of atoms in molecules (qtaim) as descriptors for use in machine learning and linear regression analyses. these were studied as an alternative to the conventionally used proxies to substituent properties. qtaim integrated group properties and local critical point properties were studied for their suitability of use in these applications. the sensitivity of these properties to the choice of density functional theory functional and basis set used for determining the density was examined. the properties at other levels of theory were compared to high-level b2plypd3-bj/aug-cc-pv5z data. integrated properties are well conserved between the different model chemistries, both in an absolute sense and in linear relationships to the reference values. local critical point properties are more sensitive to the level of theory, and hybrid functionals with triple-ζ basis sets are recommended for their evaluation. [...]