Investigando ferramentas quimiométricas de reconhecimento de padrão não supervisionado na avaliação da composição química de água mineral

Abstract

The labeling of mineral waters requires the indication of the water's chemical composition, as it may carry different denominations than “mineral water” depending on the predominant constituent ion and the source conditions. Considering the labeling process and various unsupervised pattern recognition chemometric tools, this study aimed to investigate patterns in the chemical composition of mineral waters from different sources in Brazil — specifically from the states of São Paulo and Paraná — totaling 49 samples of national origin, as well as 11 samples from foreign sources originating from France. For this purpose, Principal Component Analysis (PCA), Common Component Analysis (CCA), and Independent Component Analysis (ICA) tools were employed. Although all these tools seek to identify patterns through dimensionality reduction of the data, they differ in the manner in which this reduction is performed. The results were consistent and similar across all the analyzed tools. However, PCA and CCA showed greater agreement with each other, especially when examining the loading plots. In these plots, the samples from French sources exhibited significantly lower concentrations of sodium, fluoride, and nitrate compared to the Brazilian samples, which was expected. This is because regulations for mineral waters intended for infant consumption in France require labeling of these elements, as their excess can cause renal overload, methemoglobinemia intoxication, as well as bone, neurological, and other complications.