Utilizing visible and near infrared spectroscopy based on multi-class support vector machines classification to characterize olive oil adulteration

Abstract

Rapid and non-destructive adulteration detection is of particular importance to oil industries. This paper presents an application of visible and near-infrared spectroscopy (VNIR) for detection of adulteration levels in olive oil. Sunflower oil was used as an adulterant to olive oil and adulteration samples with different levels ranging from 0 to 40% were prepared and used for the experiments. The spectra were first considered in the range of 500-900 nm and then smoothened and normalized to reduce the light scattering effects. Principal component analysis (PCA) was performed on the spectra to have a primary data visualization and feature extraction. The extracted PCA scores were used to calculate the Mahalanobis distances of the adulterated samples from the pure sample. Further, the PCA scores were fed to the multi-class support vector machine (SVM) model to perform classification on the basis of different adulteration levels. The results showed that the spectral normalization highlighted different regions over the spectrum affected due to the adulteration. The PCA score biplots showed differences in the samples based on the different amounts of the adulteration. Moreover, the Mahalanobis distance provided a quantitative measure of the differences between the adulterated oil and the pure oil samples. The SVM modelling further supported the classification of the different levels of the adulteration. Consequently, the VNIRS in combination with the SVM could support the development of the classification protocols for detection of adulteration in olive oils.