Data evaluation for soft drink quality control using principal component analysis and back-propagation neural networks

This work describes an alternative for chemical data research, with the aim of evaluating finished product quality. Analytical data for additives in s oft drinks are interpreted by the use of multivariate data analysis: princi pal component analysis (PCA), factor analysis, cluster analysis, and artifi cial neural networks. Taking into account various chemical components like sorbic, benzoic, and ascorbic acids; saccharose; caffeine; Na, K, Ca, Mg, F e, Zn, Cu, P, and B, soft drinks were characterized and classified. The rat ios of Na, K, Ca + Mg, P, and K/Na have been studied. The application of PC A, cluster analysis, and artificial neural networks showed that combination of these chemometric tools offers effective means for modeling and classif ying soft drinks in accordance with their contents in additives and heavy m etals.