| Summary: | Extra-virgin-olive oil, honey, milk, and yogurt have associated high nutritional and commercial value. Tampering/non-conformance in these products can damage consumer's health. Therefore, rigorous quality control over the ingredients purity and declaration is necessary. The Near-infrared (NIR) is used to identify/quantify food adulterants, however the developed analytical methodologies need multivariate analysis. The portable NIR instrument enables on-site analysis, requires a few seconds, small sample volume, no sample destruction, and presents low maintenance costs. In this paper we were to classify [one-class and multi-class Support Vectors Machine (SVM), Partial Least Squares Discriminant Analysis (PLS-DA)] and PLS to quantify food adulterants using a portable NIR. The generation of artificial outliers in the one-class SVM models showed satisfactory results for authenticity analysis. The results showed that SVM (Test accuracy = 0.90-1.00) obtained better metrics compared to PLS-DA (Test accuracy = 0.83-0.97). The PLS obtained excellent accuracy: honey (RMSEP = 0.57 wt%), EVOO (RMSEP = 2.06 wt%), milk (RMSEP = 0.20 wt%), and yogurt (RMSEP = 0.06 wt%).
|
|---|