A predictive model for the sensory aroma characteristics of flue-cured tobacco based on a back-propagation neural network

Abstract

In this study, aroma quality, aroma quantity, diffusiveness and offensive odor predictions were modeled using a back propagation neural network (BPNN). The model was built from chemical components and sensory quality data features of flue-cured tobacco leaf samples. The results showed that the BPNN model adapted the optimized hidden layer well and had excellent generalization ability and robustness. The coefficients of determination (R2) for the indices of tobacco aroma characteristics based on the BPNN model, namely, aroma quality, aroma quantity, diffusiveness and offensive odor, were all above 0.7. The accuracies of the model parameters (mean squared error, regression mean standard error , and mean absolute error) were better than those of the model built using stepwise regression analysis. These results indicate that the BPNN prediction model is reliable and can accurately predict the sensory aroma characteristic qualities of flue-cured tobacco leaf.