The improvement of honey recognition models built on 1H NMR fingerprint through a new proposed approach for feature selection

Abstract

The importance of data pre-processing steps for the improvement of honey recognition models based on 1H NMR profiles was prospected and discussed in detail in the present work. These steps allowed a data dimensionality reduction, through which a very good prediction accuracy of the developed models for geographical and botanical honey differentiation was achieved. The geographical recognition models developed using the Partial Least Squares Discriminant Analysis (PLS-DA) supervised statistical method allowed a perfect classification (100 % accuracy in the cross-validation evaluation procedure) of the honey samples coming from two countries (Romania and France). For the simultaneous botanical honey discrimination of the seven varieties (acacia, linden, colza, sunflower, chestnut, lavender, honeydew) a classification power of up to 97 % was achieved.