Aroma quality characterization for Pixian broad bean paste fermentation by electronic nose combined with machine learning methods

• Published in: Journal of food measurement & characterization Vol. 18; no. 5; pp. 3363 - 3377
• Main Authors: Xu, Min, Wang, Xingbin, Xu, Zedong, Wang, Yao, Jia, Pengfei, ding, Wenwu, Dong, Shirong, Liu, Ping
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2024; Springer Nature B.V
• Subjects: Aroma; Artificial neural networks; Beans; Benzaldehyde; Broad beans; Chemistry; Chemistry and Materials Science; Chemistry/Food Science; Discriminant analysis; Electronic noses; Engineering; Fermentation; Food Science; Furfural; Guaiacol; Isopentyl alcohol; Learning algorithms; Least squares method; Machine learning; Neural networks; Original Paper; Phenylethyl alcohol; Prediction models; Support vector machines; Aroma; Pixian broad bean paste; Fermentation; Electronic nose; Machine learning
• ISSN: 2193-4126; 2193-4134
• DOI: 10.1007/s11694-024-02410-3

Pixian broad bean paste (PBP) is a popular fermentation condiment known in home and abroad. Aroma is a significant index for evaluating PBP quality during fermentation process. Hence, in this study, electronic nose (E-nose) system combined machine learning methods were applied for PBP quality characterization. The machine learning methods including partial least squares discriminant analysis (PLS-DA), partial least squares regression (PLSR), support vector machine (SVM), random forest (RF), and artificial neural networks (ANN) were introduced for qualitatively discriminating fermentation time and quantitatively analyzing the contents of key aromas of PBP samples. The PLS-DA result indicated that it is feasible to identify the fermentation stages of PBP samples by E-nose and a classification accuracy of 99% could be achieved. As for the quantitative prediction modelling, ANN exhibited preferable performance than PLSR, SVM and RF for analyzing the contents of phenethyl alcohol (R 2  = 0.846, RMSE = 10.270), isoamyl alcohol (R 2  = 0.940, RMSE = 6.857), 3-methylthiopropanol (R 2  = 0.910, RMSE = 2.205), benzaldehyde (R 2  = 0.824, RMSE = 4.172), furfural (R 2  = 0.902, RMSE = 2.066), 4-ethyl guaiacol (R 2  = 0.877, RMSE = 11.249) and 4-ethylphenol (R 2  = 0.913, RMSE = 12.754).