RAPESEED OR NOT RAPESEED? RAPID ANOMALY DETECTION USING RAMAN SPECTROSCOPY COMBINED WITH ONE CLASS CLASSIFICATION

German. Detecting anomalous samples in food and feed control is vital for ensuring consumer safety and identifying fraudulent practices. Raman spectroscopy-based fingerprinting is a rapid and non-destructive analytical method that provides an effective solution for this challenge and could be a viable alternative to systems in routine analysis. It can be used to differentiate edible oils based on their botanical origin and to detect atypical products potentially resulting from adulterations. [1, 2] This spectroscopic technique facilitates quick screening and can manage complex sample compositions. When combined with multivariate statistical analysis, it leverages the full spectral data for authentication purposes. The application of one-class classification approaches for authentication offers multiple advantages over multi-class approaches. One of the foremost advantages is that outlying samples are quickly identified and can be flagged as anomalous for further analysis. This ensures that samples with unknown adulterants can be easily detected. In this study, 138 rapeseed oils from different producers and harvest years have been analysed using a benchtop Raman spectrometer, equipped with a well plate reader, to characterize the target class. Moreover, 23 sunflower oils and 17 soybean oils were used to investigate the discrimination of the botanical origin. In order to examine the detection of adulterations, antioxidants (BHA and BHT) have been utilized as demonstrator substances to represent unknown adulterants within a concentration range of 1–10%. One-class classification successfully detected these adulterations with a specificity of 100% at concentrations of 3% and higher. This is mainly attributable to the presence of exogenous signals for BHA and BHT. In conclusion, a one-class model for verifying rapeseed oil was developed and allowed to discriminate oils of other botanical origins, to detect anomalous samples with adulterants and can thus make a contribution to consumer safety. This demonstrates the potential of Raman spectroscopy in conjunction with multivariate data analysis for the authentication of edible oils and provides the basis for future applications in routine analysis, employing the developed prediction models.