Leveraging the performance of conventional spectroscopic techniques through data fusion approaches in high-quality edible oil adulteration analyses

The high demand, high cost, and low regulations surrounding high-quality edible oils (HQEO) make them a target for fraudulent actions, particularly adulteration with refined oils. Consequently, the authentication of this kind of oil is of great interest. This work assessed the adulteration degree of five HQEOs: sesame, flaxseed, chia, rapeseed, and extra virgin olive oils, using different chemometric strategies to enhance the detection capability of the analytical methodology. Refined oils used as adulterants were evaluated at low concentrations (2–15 % v/v). Three multidimensional spectroscopic techniques (UV–Visible, near-infrared, and excitation-emission matrix fluorescence) were used, and two data fusion strategies (low- and mid-level) were evaluated. Principal component analysis was applied as an exploratory analysis tool to visualize and interpret the information contained in the dataset. For the adulterant quantification, partial least squares regression analysis was used to build the sensitive predictive models. The results revealed that chemical information enhancement leverages the ability to attain reduced prediction compared to unidimensional signals. In scenarios with low sample variability, conventional unidimensional spectroscopy (UV–Visible or near-infrared) data was shown to be adequate to guarantee predictive efficiency. In contrast, when analysing predictive figures derived from models built using a dataset with high variability, e.g., brands, low-level data fusion approaches enhance predictive efficiency. The results showed that excitation-emission matrix-based or low-level data fusion approaches can be accurately implemented to guarantee the authenticity of edible oils even when a low content of adulterant oil is presented.