Targeted multivariate adulteration detection based on fatty acid profiles and Monte Carlo one-class partial least squares

Abstract

To develop effective adulteration detection methods is essential as food quality and safety draw particular concern all over the world. In this study, Monte Carlo one-class partial least squares (MCOCPLS) was proposed and employed as a novel one class classification model for authentication identification by using virgin olive oil (VOO) as an example. Monte Carlo sampling was proposed for selecting variable subspace to improve the performance of one-class partial least squares (OCPLS) classifier. MCOCPLS was used to establish a one-class model, the performance of which was validated by an independent test set consisting of 5000 adulterated oils simulated by the Monte Carlo method. The prediction for the best model of MCOCPLS reaches a correct rate of 99.10%. Moreover, authentic VOOs were analyzed and assessed for the adulteration risk. In conclusion, the proposed MCOCPLS method could be used to effectively detect olive oils adulterated with other vegetable oils at a concentration of as low as 3%. Therefore, MCOCPLS provides an effective tool and new insights in adulteration detection for edible oils and other foods.