TastePeptides-EEG: An Ensemble Model for Umami Taste Evaluation Based on Electroencephalogram and Machine Learning

In the field of food, the sensory evaluation of food still relies on the results of manual sensory evaluation, but the results of human sensory evaluation are not universal, and there is a problem of speech fraud. This work proposed an electroencephalography (EEG)-based analysis method that effectively enables the identification of umami/non-umami substances. First, the key features were extracted using percentage conversion, standardization, and significance screening, and based on these features, the top four models were selected from 19 common binary classification algorithms as submodels. Then, the support vector machine (SVM) algorithm was used to fit the outputs of these four submodels to establish TastePeptides-EEG. The validation set of the model achieved a judgment accuracy of 90.2%, and the test set achieved a judgment accuracy of 77.8%. This study discovered the frequency change of a wave in umami taste perception and found the frequency response delay phenomenon of the F/RT/C area under umami taste stimulation for the first time. The model is published at www.tastepeptides-meta.com/TastePeptides-EEG, which is convenient for relevant researchers to speed up the analysis of umami perception and provide help for the development of the next generation of brain-computer interfaces for flavor perception.

Automated summary

This study proposes an EEG-based analysis method to effectively distinguish umami/non-umami substances. By extracting key features and using the support vector machine algorithm, the TastePeptides-EEG model is established with a validation accuracy of 90.2% and a test accuracy of 77.8%. The study discovers frequency changes and response delays in umami taste perception, providing insights for the development of brain-computer interfaces for flavor perception.