Multivariate Analysis and Classification of 146 Odor Character Descriptors

Smells can be described by assigning the words that come to mind when sniffing an odorous material. A great number of terms can be applied, but not all of them are independent, and it is possible to establish groups of words often applied together when describing a smell. Such classification of olfactory descriptors is of scientific interest in order to better understand the dimensionality and structure of human olfactory perception space. For this purpose, compilations of olfactory profiles contain valuable information that may lead to certain consensus in odor classification. One of the most comprehensive odor databases is the Dravnieks’ Atlas, which contains quantitative olfactory profiles for 160 samples. For each one, a large panel rated the applicability of 146 odor character descriptors on a numeric scale. By applying principal component analysis to this Atlas, 105 descriptors were reorganized in 24 classes, and 33 attributes were considered as odors intermediate of two or three categories. The similarities between classes were studied by means of a further multivariate analysis based on latent variables, which provides valuable information about the most salient dimensions of odor space. Consistent with other reported statistical analyses of olfactory databases, the perceptual space of odor character is multidimensional with about 20–30 dimensions, and it is better described as a continuum spectrum rather than as a segmented space. Attempts to classify all possible odor descriptors in a restricted number of classes appear to be inappropriate. Instead, 24 categories of related terms are proposed here, regarding the rest as intermediate smells, assuming that olfactory classes are not independent and follow certain hierarchy according to particular underlying dimensions.