UV-Visible Spectroscopy and Multivariate Classification as a Screening Tool for Determining the Adulteration of Sauces

Abstract

A rapid, simple, and economic multivariate screening methodology based on UV-visible spectroscopy and multivariate classification is proposed to test for adulteration in sauces. Two classification strategies were evaluated to compare their ability to detect food fraud: untargeted modeling (class modeling) and targeted classification (discriminant analysis). As a case study, the possible adulteration of ketchups and barbecue sauces with the banned Sudan I dye was considered. The classification models were built with a new classification tool for class modeling (partial least squares-density modeling, PLS-DM) and with the classical discriminant partial least squares-discriminant analysis (PLS-DA). Very satisfactory classification results were obtained with both strategies: regarding untargeted modeling, only original samples (class 1) were modeled obtaining a 94.5 % of correct classification and regarding targeted classification, two classes were considered (class 1 original samples and class 2 adulterated samples) with an overall classification rate of 97.3 %. The two strategies are useful and adequate as screening tools for monitoring the quality of sauces especially in situations that require quick responses.