Enhanced Specificity for Detection of Frauds by Fusion of Multi-class and One-Class Partial Least Squares Discriminant Analysis: Geographical Origins of Chinese Shiitake Mushroom

Abstract

Both multi-class and one-class discrimination analyses (DAs) have been widely used for tracing the geographical origins of Protected Designation of Origin (PDO) foods. However, due to the complexity of potential non-PDO frauds, both methods have encountered some problems. Because multi-class DA tries to classify two or more predefined classes, its classification results will be unreliable when it is used to predict a new object from an untrained class. One-class DA is developed using only the information concerning one-class objects, so they cannot necessarily ensure the model specificity for detection of various food frauds. In this work, a new chemometric strategy was proposed by fusion of multi-class and one-class DA to trace the geographical origin of a Chinese dried shiitake mushroom with PDO. The PDO shiitake objects (n = 161) and non-PDO objects (n = 264) from five other main producing areas were analyzed using near-infrared spectroscopy. The classification performance of multi-class DA, one-class DA, and model fusion was compared. With second-order derivative (D2) spectra, model fusion obtained a high sensitivity (0.944) and specificity (0.968). Model comparison indicates that fusion of multi-class and one-class DA can enhance the specificity for detecting various non-PDO foods with little loss of model sensitivity.