Comparative analysis of feature extraction methods in the clustering of electronic nose response correlated with GC/MS analysis

Abstract

This paper discusses feature extraction of electronic nose (e-nose) responses correlated with gas chromatography-mass spectrometry (GC/MS) analysis. Seven kinds of herbal essential oils were measured by e-nose, including ginger, lesser galangal, turmeric, zedoary, greater galangal, patchouli, and wild ginger. Three methods, i.e. relative amplitude (RA), surface (SF) and 3rd scale wavelet decomposition (WD) were compared to extract e-nose responses. For this purpose, two common clustering methods, k-means algorithm (KMA) and hierarchical agglomerative algorithm (HAA), were used to cluster the response dataset extracted using the RA, S and WD method. Two external measures of cluster validity, e.g. purity (P) and rand index (RI) were used to evaluate the quality of the clustering on 35 samples of herbal essential oils extracted by the RA, SF and WD method. Experiment results show that RA method obtains the best quality in k-means clustering, as well as to cluster 7 herbal essential oils using HAA and correlated with GC/MS analysis result.