Enhancing classification rate of electronic nose system and piecewise feature extraction method to classify black tea with superior quality
Kombo Othman Kombo, Nasrul Ihsan, Tri Siswandi Syahputra, Shidiq Nur Hidayat, Mayumi Puspita, Wahyono Wahyono, Roto Roto, Kuwat Trıyana
Abstract
This study introduced a metal-oxide-semiconductor (MOS) based electronic nose (E-nose) to perform on-the-spot classification of superior-quality black tea. A piecewise feature method based on a line-fitting model was introduced to extract comprehensive features of E-nose sensor response curves. Principal component analysis (PCA) and linear discriminant analysis (LDA) were used for data dimensionality reduction and structure visualization. Support vector machine (SVM) with a Radial kernel function was used to assess the performance of E-nose. The results indicated that the SVM model coupled with the piecewise feature method performed better and achieved the best classification rates of 99.50%, 95.30%, and 96.50%, for training, validation, and testing datasets respectively, with testing sensitivity and specificity of up to 98.6% and 99.10%. The E-nose result was further correlated with compound concentrations in the black tea, measured using gas chromatography-mass spectrometry (GC-MS). Based on its enhanced performance evaluation, the introduced lab-built E-nose system yielded promising results in assessing superior-quality black tea.