Counteracting drift of olfactory sensors by appropriately selecting features

It is shown that the effects of olfactory sensor drift can be counteracted by appropriately selecting the features that characterise the sensor responses. To this end, a supervised version of the fuzzy isodata (SFI) algorithm is adopted. In addition to selecting features, the SFI algorithm computes both the memberships of patterns in classes, and the shape of classes. The output of the SFI is then used by a fuzzy k-nearest neighbour algorithm to identify unknown odours