Chemometrics in Food Authentication

The guarantee of food safety and quality along the food chain is demanding great attention from the food industry and consumers. Regulatory authorities, retailers, food producers, and processors are all actively interested in ensuring that food quality matches by the claims they are posing. Fraud and adulteration detection, so-called food authentication, is a process that verifies food compliance with its label description [1]. This may include the geographical origin, production method, processing technologies, and food composition. Various types of food fraud, including substitution, dilution, tampering, misrepresentation, and mislabeling, are often used to derive financial profits by reducing manufacturing costs or extending shelf lives. Authenticating foodstuffs that hold a certificate of specific character (CSC), protected designation of origin (PDO), or protected geographical indication (PGI) is essential with agro-food products [2]. Food adulteration is a growing challenge for food manufacturers and analysts because most adulterants are unknown and difficult to recognize using the typical targeted screening methods. This industry urgently needs non-targeted screening methods for food samples to provide proof of origin and prevent deliberate or accidental undeclared additions to those samples. Nowadays, the rapid growth of the capabilities of modern analytical instruments provides a large amount of data involving a wide range of factors (features) that need chemometrics for the extraction of meaningful information. Chemometrics provides powerful tools in calibration and classification analysis, applied in targeted and non-targeted approaches to identify various food fraud situations or verify of their geographic or biological origin. The most common multivariate methods and principles for food authentication can be classified into three categories: exploratory data analysis; data description and visualization, discrimination, and classification; and regression and prediction. Classification methods based on multivariate data analysis could be supervised or unsupervised. In unsupervised methods, the purpose is to identify clusters or relationships between samples without prior knowledge of classes or groups. In contrast, supervised methods require information on class membership and a training stage to build a proper mathematical model. Classification techniques are of either of two main types. One is known as “discriminant analysis” or “hard modeling” and partitions the data space into isolated regions (classes); the other, known as “class modeling” or “soft modeling”, models each category or class separately [3].