Incorporation of two-dimensional correlation analysis into discriminant analysis as a potential tool for improving discrimination accuracy: Near-infrared spectroscopic discrimination of adulterated olive oils

Abstract

A strategy of combining temperature-induced spectral variation and two-dimensional correlation (2D-COS) analysis as a potential tool to improve accuracy of sample discrimination is suggested. The potential application of this method was evaluated using near-infrared (NIR) spectroscopic discrimination of adulterated olive oils. Rather than utilizing static spectral information at a certain temperature, dynamic spectral features induced by an external perturbation such as temperature change would be more informative for sample discrimination, and 2D-COS analysis was a reliable choice to characterize temperature-induced spectral variation. For evaluation, NIR spectra of 9 pure olive oils and 90 olive oils adulterated with canola, soybean, and corn oils (adulteration rate: 5%) were collected at four different temperatures (20, 27, 34, 41 °C). In constant-temperature measurements, the scores of pure and adulterated samples obtained by principal component analysis (PCA) were considerably overlapped. When 2D-COS analysis was performed using temperature-varied (20–41 °C) spectra and the resulting power spectra from 2D synchronous correlation spectra were used for PCA, identification of the two groups was noticeably enhanced and subsequent k-nearest neighbor (k-NN)-based discrimination accuracy substantially improved to 86.4%. While, the accuracies resulted in the constant-temperature measurements ranged only from 50.9 to 55.8%. The dynamic temperature-induced spectral variation of the samples effectively featured by 2D-COS analysis was ultimately more informative and allowed improvement in accuracy.