A genetic algorithm (GA) combined with partial least squares (PLS) regression was applied in order to reduce the considerable number of variables of electrochemical signals recorded by a voltammetric electronic tongue. The algorithm was specifically designed for improving the analysis of data matrices with high collinearity, providing good solutions in terms of both predictive ability and interpretability and, at the same time, minimising the risk of overfitting.
The variable reduction was carried out within a wine characterisation performed with an electronic tongue based on voltammetric sensors. In more detail, the aim was the evaluation of differences among wines aged by means of two different processes: the traditional ageing process in oak wood barrels and an alternative method based on the use of stainless steel tanks and oak wood chips or staves. The results show that the variables selected are at least as good as the full signals to separate the two different classes, even though the number of parameters has been reduced from 1980 to 280 variables; furthermore, the models obtained are simpler and more easily interpretable. In fact, it is important to point out that the variables selected by the genetic algorithm are related to the electrochemical activity of the polyphenolic fraction of wine.
Excepté là où spécifié autrement, la license de ce document est décrite en tant que Attribution-NonCommercial-NoDerivatives 4.0 International