Non-destructive determination of volatile compounds and prediction of amino acid nitrogen during sufu fermentation via electronic nose in combination with machine learning approaches

Abstract

Electronic nose along with different machine learning approaches was tested as tool to predict degree of sufu fermentation, with amino acid nitrogen content as marker. Consistent increase in E-nose signal values and amino acid nitrogen content were noted with fermentation time. The number of volatile compounds in sufu increased from 19 to 61 upon fermentation, with 46 compounds identified as major aroma contributors. Among them, acetone, ethyl palmitate and indole demonstrated strong correlations with amino acid nitrogen content. Three prediction models, i.e., partial least-squares regression, support vector machine, and artificial neural network, were tested for abilities to predict amino acid nitrogen content. All models were capable of quantitatively predicting the content of amino acid nitrogen, with ANN-based model demonstrating the highest prediction accuracy (RP2 = 0.968, RMSE = 5.49 × 10−8). Electronic nose along with appropriate prediction model is noted as a promising tool for evaluating the degree of sufu fermentation.