**6. Conclusions**

For unifloral honey samples classification, a similar pattern recognition artificial neural network was built, with 6 neurons in the input layer, 3 neurons in the outer layer, and 10 neurons in the hidden layer. A total of 70% of the 270 unifloral honey samples were used for training, 15% for testing, and 15% for validation. The best results obtained led to a correct group assignment with a total error of only 3.3%. For each honey type, the errors in the sample rec-

ognition were: 4.4% for acacia, 5.6% for linden, and 0% for colza (**Figure 15**).

56 Honey Analysis

**Figure 15.** Confusion matrix for unifloral samples classification (1–acacia, 2–linden, 3–colza).

The complexity of honey characterization, control, and classification has been presented using a large pool of scientific evidence, brought in by many Romanian researchers. Compared to the honey from other European countries, the Romanian honey has good market qualities due to its organic character and various botanic sources responsible for the specific flavour and consistency. The original case study presented confirms the possibility of discrimination between different honey types, based only on physico-chemical properties measurements, as demanded by the quality control.
