Application of stable isotopic and elemental composition combined with random forest algorithm for the botanical classification of Chinese honey

To ensure that honey belongs to a very appreciated botanical class, the classical methodology is melissopalynology analysis to identify and count pollen grains. However, this method is time-consuming and laborious. In this work, four stable isotopes (δ13Ch, δ13Cp, δ18O and δ2H) and twelve elemental contents (Na, Mg, Ca, K, Fe, Cr, Mn, Co, Cu, Sr, Se, Mo) were used to build the dataset, and the Random Forest (RF) algorithm, Support Vector Machines (SVM), Classification and Regression Trees (CART) and Linear Discriminant Analysis (LDA) were investigated to classify six varieties of Chinese honey (linden, sunflower, vetch, rape, acacia, and jujube). The results showed the RF algorithm exhibits the highest training accuracy (99.4%) and testing accuracy (96.5%) of the four algorithms. Hence, the RF algorithm was selected to rank the 16 attributes according to their contribution, and δ2H, Sr, δ18O, Mn, Ca, and K were considered the most important factors for identifying six varieties of honey. Furthermore, the results of the RF algorithm were verified by the parallel coordinates plot. This suggests that the RF algorithm provides an effective and accurate approach for classifying Chinese honey according to stable isotopic and elemental composition, which theoretically can be used to classify more types of honey.