Develop of a Sample Classifier Through Multivariate Analysis for Caffeine as a Bitter Taste Generator

Abstract

The use of voltammetric electronic tongues (VET) in the food industry is becoming more frequent, being used both for the identification of substances such as antioxidants or phenols, as well as in the case of sensory analysis simulation applications to help with quality control analysis or development of new products. In this work VET developed by the IDM of the UPV were employed, to get the data and then classify caffeine with different levels of intensity. The data was processed by multiclass analysis through a supervised learning algorithm which uses a vector support machine, with binary learners using the one-versus-all coding design, choosing a linear function as classifying element, the aim of this work was to verify if the VET can differentiate between different samples of caffeine which is the main reference for the bitter flavor. The results showed a concordance of 67.19% in the separation of samples, allowing to conclude as regular the performance of the classifier and therefore the data acquired through the VET