Application of least squares support vector machines for discrimination of red wine using visible and near infrared spectroscopy

Visible and near infrared (Vis/NIR) transmittance spectroscopy and chemometrics methods were utilized to discriminate red wine. The samples of five varieties of red wine were separated into calibration set and validation set randomly. The principal components (PCs) could be obtained from original spectrum by using Partial least squares (PLS), The PCs (selected by PLS) of each sample in calibration set was used as the inputs to train the Least squares support vector machines (LS-SVM) model, then the optimal model was used to predict the varieties of samples in validation set based on their PCs, and 94% recognition ratio was achieved with the threshold predictive error ±0.1, while 100% recognition ration with the threshold predictive error ±0.2. Root mean square error of prediction (RMSEP) and determination coefficient (r²) were 0.0531 and 0.9986 respectively. It is indicated that Vis/NIR transmittance spectroscopy combined with PLS and LS-SVM is an efficient measurement to discriminate types of red wine.