Análise multivariada e estratégia de fusão de dados na detecção e caracterização de uísques falsificados

Abstract

Cases of fraud in alcoholic beverages are common, demanding to be specific to consumers, can affect the health of those who drink a fraudulent drink, cause great damage to original brands, and drive the criminal economy. In the forensic context, the search for quick and low-cost analysis strategies has been a routine work of public security professionals to reduce cases of crimes involving the trade of counterfeit alcoholic beverages and to contain the advancement of procedures that lead to the sophistication of fraud. This work aimed to develop multivariate analysis methodologies associated with three analytical techniques (NIR spectroscopy, ICP MS, and physicochemical parameters) to detect and characterize fraud in whiskeys. In this way, seven classification models were constructed using the PLS-DA method, three containing the measurements of each analytical technique, and four data fusion models based on physicochemical, atomic, and molecular measurements. The models were evaluated according to their ability to discriminate between authentic and counterfeit samples of whiskey seized mainly by the Civil Police of the state of Minas Gerais (PCMG). The performance of the models was evaluated using figures of merit such as sensitivity (SEN), specificity (SPE), false negative rate (TFN), false positive rate (TFP), and reliability rate (TCONF). The models were characterized using information, VIP scores, and regression vectors. The results demonstrated that all the models built had predictive capacity in the test sets equal to 100% in the SEN, SPE and TCONF criteria, while the TFN and TFP criteria were equal to zero. The characterization of the models showed that the ethanol content, color, and copper content are the main variables that characterize authentic whiskey samples. The models developed have the potential to be inserted into the work routine of PCMG's technical expert staff.