Abstract

Botanical differentiation of four floral honey types based on isotopic and elemental content coupled with partial least squares—discriminant analysis (PLS-DA) was performed in the present work. In order to enhance the discrimination potential of this approach, an advanced pre-processing step was applied. For data pre-processing, a comparison between unsupervised (principal component analysis - PCA) and supervised (PLS-DA) methods was performed to achieve the highest classification rate of the developed models. Based on the prediction rate of the final models, it was assumed that the most efficient data pre-processing method corresponds to the selection of relevant features by PLS-DA. The statistical models allowed the differentiation of individual classes with percentages between 82% and 100%. Thus, the model accuracies for the recognition of each honey category were 100% (sunflower), 92% (acacia), 90% (colza) and 82% (linden). The applied variable selection, performed through PLS-DA, led to a significant improvement of the models in the cross-validation evaluation (i.e., for the sunflower honey samples, the true positive rate increased from 66% to 83%).

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call