Comparative Analysis of Feature Extraction Techniques for Authentification of Brand Perfume using Electronic Nose
Adhytia Tungga, Riser Fahdiran, and Bambang Heru Iswanto

Complex Systems Research Group, Department of Physics, Universitas Negeri Jakarta, Jl. Rawamangun Muka, Jakarta 13220, Indonesia

Abstract

Feature extraction is an important step in perfume identification using electronic nose (e-nose) efficiently. In this paper, we present a comparison of several feature extraction techniques under different clustering methods. The sample used was an authentic and six fake perfume samples. The aim of this study is to show the relevant feature extraction technique that improves the cluster validity. The feature extraction methods studied are: relative amplitude (RA), surface (S), spectral centroid (SC), skewness (SK), wavelet decomposition (WD), and wavelet entropy (WE). Data analysis uses the principal component analysis (PCA) method and general clustering algorithms, such as k-Means, agglomerative hierarchical clustering (AHC), and Gaussian mixture models (GMM). Cluster validity on e-nose response data is measured using purity and entropy. The results obtained confirmed the legitimacy of the electronic nose technique as an alternative to the sensory analysis as far as the determination of authenticity of perfume is concerned.

Keywords: feature selection, electronic nose, clustering, perfume scent

Topic: Computer Science