Interpretable support vector machine for authentication of omega-3 fish oil supplements using Raman spectroscopy

The use of machine learning algorithms to develop automated analytical methods and smart sensors has drastically increased in recent years. Although these algorithms often provide great performances, the poor interpretability is one of their main limitations. In this work, we present an efficient method to authenticate certified omega-3 fish oil supplements which relies on a vibrational spectroscopic technique (Raman spectroscopy) and a chemically interpretable machine learning algorithm (interval support vector machine, iSVM). The Raman spectra were collected from 248 certified and 520 non-certified omega-3 fish oil samples in a non-destructive fashion. Key differences in the chemical composition of certified and non-certified supplements were unveiled by correlating the iSVM discriminant capability with Raman bands in the corresponding spectral intervals. Proper figures of merit (e.g., accuracy and F1 score) were estimated and critically discussed, and an additional external validation was performed with independent samples to assess the model's robustness and generalization capability. In addition to the excellent chemical interpretability, which revealed the great importance of Raman bands of unsaturated fats to authenticate samples, the proposed Raman-iSVM method showed to be accurate (96 and 92% for internal and external validation sets, respectively), rapid, portable, and computationally costeffective. Therefore, we believe this method can readily be implemented for in-situ quality monitoring of omega-3 fish oil supplements and other food products.