Characterization of cognacs and grape brandies by fluorescence spectra processed using machine learning methods

Abstract

A method for express characterization of cognacs and grape brandies is proposed in the case study of their classification by geographical origin. The method is based on the use of informative fragments of fluorescence spectra of samples of different geographic origin and their subsequent processing using machine learning algorithms. Three types of fluorescence spectra were selected, i.e., spectra of synchronous scanning at a wavelength difference of 50 nm, and emission spectra at an excitation wavelength of 250 and 280 nm. These spectra were measured for 43 samples of cognacs and grape brandies, which were divided into 3 classes according to their geographical origin, the regions of the Russian Federation (except for Dagestan), the Republic of Dagestan (Russian Federation), and the Republic of Armenia. A training set consisting of 33 samples and a test set consisting of 10 samples were formed from the samples under study. To train the models, an extreme gradient boosting, one of the modern machine learning algorithms, was chosen as suitable for a limited number of samples in the training set. The correctness of the sample recognition of the test set (consisting of 10 samples not used in training) was 100% for models based on emission spectra and spectra of synchronous scanning. The results obtained demonstrate the fundamental possibility of using informative fragments of fluorescence spectra in combination with machine learning to characterize cognacs and grape brandies, including their classification by the geographical origin. However, the use of this method in regulated procedures of the product control is possible only for cognacs and grape brandies with a protected geographical indication (designation of the origin). The above approach can also be used to classify other liquid food products (juices, honey, etc.).