Nonlinear canonical correspondence analysis: description of the data of coffee

Abstract

The formulation of coffee blends is of paramount importance for the coffee industry, as it provides the product with an expressive ability to compete in the market and adds sensory attributes that complement the consumption experience. Through redundancy analysis and canonical correspondence analysis, it is possible to study the relationships between a set of sensory notes and a set of blends with different proportions of coffee variety through multivariate linear regression models. However, it is unrealistic to assume that such sensory responses are given linearly in relation to the formulation of the blends, since some coffee species have greater weight in the sensory evaluation (quadratic terms) and the effect of the mixtures (term of interaction). With this motivation, this work aims to propose the use of redundancy analysis and nonlinear correspondence analysis through multivariate polynomial regression to evaluate the acceptance of different varieties of coffee blends according to the scores given by the evaluators. Finally, it is concluded that there were gains in the percentage of total explained variance in the polynomial models in relation to the classic models.