Indoor Localisation using Aroma Fingerprints: Comparing Nearest Neighbour Classification Accuracy using Different Distance Measures

Abstract

Measurements from an ion mobility spectrometry electronic nose (eNose) can be used for distinguishing different rooms in indoor localisation. An earlier study showed that the Nearest Neighbour classifier with Euclidean distance for features provides reasonable accuracy under certain conditions. In this paper 66 alternative distance measures are compared to the Euclidean distance and principal component analysis (PCA) is applied to the data. PCA shows that the measurements on the various channels of the eNose are correlated and that using principal components 1, 2 and 4 increases the accuracy considerably. Furthermore, the experiments revealed three Pareto optimal distance measures that reduce the misclassification rate between 9-10% while using only 82-88% of the search time compared with Euclidean distance.