Comparison of Multivariable Techniques for Brand Classification of Turmeric Powders by Near-infrared (NIR) Spectroscopy

In this work, near-infrared (NIR) spectroscopy was applied for the rapid classification of 7 different brands of turmeric powder. Multivariate calibrations including nearest neighbor (kNN) and support vector machine (SVM) were studied and compared to achieve an optimal result. At first, to visualize the different clustering of the turmeric brands principal component analysis (PCA) was performed to visualize and principal components (PCs) were derived which were applied as inputs to the kNN and SVM models. Cross validation was used to extract number of PCs and optimization parameters. The reflectance spectra of 84 total samples were taken by NIR spectrometer for the analysis purpose out of which 70% was used for the training purpose and 30% for testing purpose. Figures of merit (FOM) were extracted from the confusion matrices of the kNN and SVM classifiers. Comparing the two investigated models, both kNN model and SVM model effectively classified all samples in test set according to their brands with accuracy ranging from 0.76-0.91% and 0.79-0.95% respectively.