Chemometrics as an efficient tool for food authentication: Golden pillars for building reliable models

Detecting food fraud or confirming the authenticity, which falls within the general concept of food integrity, is a complex problem. Modern analytics platforms are used to address these issues, and multivariate data analysis techniques help to extract important information from the signals generated. The main chemometric/machine-learning methods for solving authentication problems are one-class classifiers (OCC), whose goal is to detach a class of genuine/pure/non-adulterated samples from all other samples and classes, by capturing the main similarities within the samples of the target class. Basic concepts and new trends of one-class classifiers are discussed together. The special case of authentication tasks, where a small amount of illegal ingredients can significantly affect the quality of the product, is also considered. Such cases demand estimation of the limits of detection of non-acceptable ingredients as a final solution. Modern trends, such as hierarchical modeling, multi-platform analytical approaches and hyperspectral imaging, together with examples of successful applications for food authentications are discussed. An overview of commercial and free software packages is provided for practical applications. Many chemometric methods are available for solving various food authentication tasks via user-friendly software packages. Regardless of how the task is set and how informative the fingerprints are, some analytical results are always obtained. In order to build reliable models and to obtain interpretable results for further routine analyses, it is recommended to adhere of ten basic principles presented at the end of the review.