LAUSR

Abstract The lipid chromatographic fingerprint of different avocado fruits have been acquired and two classification multivariate methods, partial least squares-discriminant analysis (PLS-DA) and support vector machine (SVM), have been successfully tested in order to discriminate and classify a higher variability of avocado samples. Two authentication goals have been achieved attending to: (i) the geographical origin, and (ii) the botanical variety or cultivar. However, to our knowledge, there are no antecedents aimed at comparing and classifying avocado fruits. The pulp oil fraction of the avocado fruit was first extracted using pressurised liquid extraction from the previously lyophilised pulp. Then the 190–400 ​nm UV-absorption fingerprints were obtained from the avocado oils using normal phase high performance liquid chromatography coupled to an absorption diode-array detector ((NP)HPLC-DAD) and the 220 ​nm spectra were then selected for classification model building. Several input-class classification strategies were applied and the classification models were externally validated from the specific success/error contingencies. In addition, some quality metrics, i.e. sensitivity (or recall), specificity, precision, negative predictive values, efficiency (or accuracy), AUC (area under the receiver operating curve), Mathews correlation coefficient and Kappa coefficient, were determined to evaluate the performance of each classification model (PLS-DA and SVM) and the results clearly show that SVM method is the most proficient.