Predicting the Botanical Origin of Honeys with Chemometric Analysis According to Their Antioxidant and Physicochemical Properties

The aim of this study was to develop models based on Linear Discriminant Analysis (LDA), Classification and Regression Trees (C&RT), and Artificial Neural Network (ANN) for the prediction of the botanical origin of honeys using their physicochemical parameters as well as their antioxidative and thermal properties. Also Principal Component Analysis (PCA) and Cluster Analysis (CA) were performed as initial steps of data mining. The datasets consisted of 72 honey samples (false acacia, rape, buckwheat, honeydew, linden, nectar-honeydew and multi oral) obtained from different regions of Poland and collected between April 2014 and November 2016. Ash content, pH, free acidity, colorimetric coordinates in the CIELAB space (L*, a*, b*, h*, C*), total phenolics content, antioxidant activity, and glass transition temperatures (T-g) of the honey samples were determined. The first four principal components accounted for about 85% of the total variance. PC1 was highly correlated with colour intensity, the hue angle (h*), and total phenolics content, whereas PC2 was dominated by chroma (C*) value and glass transition temperatures (T-g). The CA dendrogram displays two clusters: one with light coloured honey samples and second with dark coloured honey samples. On the basis of the LDA analysis, the colour parameters possessed the highest discrimination power according to the botanical origin of honey samples. The models based on ANN and C&RT algorithms were characterized by 100% accuracy. Study results demonstrate that the chemometric approach enables high-accuracy classification of honeys according to their botanical origin.