Abstract

This work presents the development and application of a method based on synchronous fluorescence coupled with chemometric tools to classify different vodka samples. The chemometric methods used encompasses partial least squares – discriminant analysis (PLS-DA), k-nearest neighbor (KNN), and support vector machine (SVM). A total of 18 authentic vodkas (3 brands, 6 of each brand, all donated by the producers) and six counterfeit vodka samples were used to validate the method. The samples suspect of falsification were seized in a police action that has occurred in the state of São Paulo, Brazil named Operation Chicago. The spectrofluorescence dataset was processing using the centering on the mean before to apply principal component analysis (PCA), which did not correctly discriminate the samples. Considering that the PCA suffers from the presence of outliers, robust PCA was applied, which detected outliers. After this undesirable detection, the transformation of the spatial signal was applied, and the robust PCA has applied again and did not detect outliers. The PLS-DA and SVM were able to classify all the vodka samples correctly in authentic and counterfeit. Both methods showed the highest values for the accuracy and Kappa index, as well the sensitivity and specificity, respectively. However, the KNN has not been able to correctly classify the samples. Finally, only the SVM based on radial base function was able to classify all brands of vodka samples correctly using synchronous fluorescence spectroscopy.

摘要

这项工作介绍了基于同步荧光结合化学计量学工具对不同伏特加样品进行分类的方法的开发和应用。使用的化学计量学方法包括偏最小二乘-判别分析（PLS-DA），k最近邻（KNN）和支持向量机（SVM）。该方法总共使用了18种正宗伏特加酒（3个品牌，每个品牌6个，均由生产商捐赠）和6个假冒伏特加酒样品。在巴西圣保罗州发生的一项名为“芝加哥行动”的警察行动中，查获了伪造样品的嫌疑人。在应用主成分分析（PCA）之前，以平均值为中心对光谱荧光数据集进行了处理，该方法无法正确地区分样品。考虑到PCA遭受异常值的影响，因此应用了鲁棒的PCA来检测异常值。在执行此不良检测之后，将应用空间信号的变换，并且将鲁棒PCA再次应用，并且不会检测到异常值。PLS-DA和SVM能够正确地将所有伏特加样品分类为真品和伪造品。两种方法均显示出最高的准确性和Kappa指数值，以及敏感性和特异性。但是，KNN无法正确分类样本。最后，只有基于径向基函数的SVM才能使用同步荧光光谱法对所有品牌的伏特加样品进行正确分类。应用了空间信号的变换后，鲁棒的PCA又被应用了，并且没有检测到异常值。PLS-DA和SVM能够正确地将所有伏特加样品分类为真品和伪造品。两种方法均显示出最高的准确性和Kappa指数值，以及敏感性和特异性。但是，KNN无法正确分类样本。最后，只有基于径向基函数的SVM才能使用同步荧光光谱法对所有品牌的伏特加样品进行正确分类。应用了空间信号的变换后，鲁棒的PCA又被应用了，并且没有检测到异常值。PLS-DA和SVM能够正确地将所有伏特加样品分类为真品和伪造品。两种方法均显示出最高的准确性和Kappa指数值，以及敏感性和特异性。但是，KNN无法正确分类样本。最后，只有基于径向基函数的SVM才能使用同步荧光光谱法对所有品牌的伏特加样品进行正确分类。以及敏感性和特异性。但是，KNN无法正确分类样本。最后，只有基于径向基函数的SVM才能使用同步荧光光谱法对所有品牌的伏特加样品进行正确分类。以及敏感性和特异性。但是，KNN无法正确分类样本。最后，只有基于径向基函数的SVM才能使用同步荧光光谱法对所有品牌的伏特加样品进行正确分类。