Assessing the ability to predict human percepts of odor quality from the detector responses of a conducting polymer composite-based electronic nose

The responses of a conducting polymer composite "electronic nose" detector array were used to predict human perceptual descriptors of odor quality for a selected test set of analytes. The single-component odorants investigate d in this work included molecules that are chemically quite distinct from e ach other, as well as molecules that are chemically similar to each other b ut which are perceived as having distinct odor qualities by humans. Each an alyte produced a different, characteristic response pattern on the electron ic nose array, with the signal strength on each detector reflecting the rel ative binding of the odorant into the various conducting polymer composites of the detector array. A "human perceptual space" was defined by reference to English language descriptors that are frequently used to describe odors . Data analysis techniques, including standard regression, nearest-neighbor prediction, principal components regression, partial least squares regress ion, and feature subset selection, were then used to determine mappings fro m electronic nose measurements to this human perceptual space. The effectiv eness of the derived mappings was evaluated by comparison with average huma n perceptual data published by Dravnieks. For specific descriptors, some mo dels provided cross-validated predictions that correlated well with the hum an data (above the 0.60 level), but none of the models could accurately pre dict the human values for more than a few descriptors. (C) 2001 Elsevier Sc ience B.V. All rights reserved.