Torres Mariani, Nathalia CristinaCosta, Rosangela Camara daGomes de Lima, Kassio MichellNardini, VivianiCunha Junior, Luis CarlosAlmeida Teixeira, Gustavo Henrique de [UNESP]2015-03-182015-03-182014-09-15Food Chemistry. Oxford: Elsevier Sci Ltd, v. 159, p. 458-462, 2014.0308-8146http://hdl.handle.net/11449/116549The aim of this study was to evaluate the potential of near-infrared reflectance spectroscopy (NIR) as a rapid and non-destructive method to determine soluble solid content (SSC) in intact jaboticaba [Myrciaria jaboticaba (Veil.) O. Berg] fruit. Multivariate calibration techniques were compared with pre-processed data and variable selection algorithms, such as partial least squares (PLS), interval partial least squares (iPLS), a genetic algorithm (GA), a successive projections algorithm (SPA) and nonlinear techniques (BP-ANN, back propagation of artificial neural networks; LS-SVM, least squares support vector machine) were applied to building the calibration models. The PLS model produced prediction accuracy (R-2 = 0.71, RMSEP = 1.33 degrees Brix, and RPD = 1.65) while the BP-ANN model (R-2 = 0.68, RMSEM = 1.20 degrees Brix, and RPD = 1.83) and LS-SVM models achieved lower performance metrics (R-2 = 0.44, RMSEP = 1.89 degrees Brix, and RPD = 1.16). This study was the first attempt to use NIR spectroscopy as a non-destructive method to determine SSC jaboticaba fruit. (C) 2014 Elsevier Ltd. All rights reserved.458-462engNIR spectroscopyPLSBP-ANNLS-SVMVariables selectionArtigo10.1016/j.foodchem.2014.03.066WOS:000336109500065Acesso restrito