Classificação de iogurte quanto à presença de lactose por meio da espectoscopia NIR e quimiometria

• Ano de defesa: 2024
• Autor(a) principal: Silva, José Manuel Amancio da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Química; Programa de Pós-Graduação em Química; UFPB
• Palavras-chave em Português: Iogurte; Lactose; Espectroscopia NIR; Yogurt; Spectroscopy NIR; BA-LDA; GA-LDA; SPA-LDA; SIMCA; CNPQ::CIENCIAS EXATAS E DA TERRA::QUIMICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/35143

Yogurt is a nutrient-rich beverage and an excellent source of calcium, proteins, and vitamins, produced through lactic fermentation by microorganisms. Yogurt can be classified as a functional food due to its nutritional properties that support the proper functioning of the digestive system. Given that lactose intolerance affects over 75% of the global population, the enzyme β-galactosidase is used industrially to reduce lactose content in milk and its derivatives. Quality control of food typically relies on conventional analytical techniques that are time-consuming and have high operational costs. Therefore, it is necessary to develop alternative techniques that are rapid, reagent-efficient, and allow for in situ measurements. Hence, there is growing interest in alternatives to conventional methods, such as using Near-Infrared Spectroscopy (NIR) in combination with chemometrics. The purpose of this study was to propose a non-destructive, rapid, and cost-effective analytical strategy— using NIR spectrometry technique, combined with chemometric multivariate analysis tools — to classify yogurt according to lactose presence. For this, a total of 102 samples of zero and regular yogurt were acquired. Diffuse reflectance spectra, obtained in the range of 900 to 1700 nm, were collected using portable NIR equipment. Various pre-processing techniques were evaluated, as well as the use of cross-validation and the classification models Linear Discriminant Analysis (LDA), with selection of variables by Bat Algorithm (BA), Genétic Algorithm (GA) and Successive Projections Algorithm (SPA), and the Soft Independent Modeling of Class Analogy (SIMCA). The classification models with BA-LDA and GA-LDA variable selection correctly classified all samples, achieving maximum performance values. Only the SPA-LDA models with raw data and SNV pre-processing did not reach 100% performance for the training set. SIMCA class modeling showed overlap between PCA models of the classes, resulting in misclassification of samples and lower performance values. The results presented in this study indicate that the proposed strategy, combining NIR spectroscopy and multivariate analysis, proved to be a viable alternative tool for the classification of lactose-containing and lactose-free yogurts, particularly using LDA modeling.