Data Fusion Models for Assessing Adulteration in Butter Through Its Vibrational Spectroscopic Fingerprint

Food adulteration is a longstanding issue that significantly impacts consumers’ health and well-being on a small scale, while affecting the economy as a whole on a larger scale. This practice is becoming increasingly sophisticated every day and has seen a notable rise in recent years, especially after the COVID-19 pandemic. Even more concerning, today, any food item is vulnerable to adulteration. This fact, among other reasons, is why controlling fraudulent food requires a multifaceted approach, including public policies, comprehensive consumer information, and the development of detection methodologies. As dairy products are highly adulterated overseas, various instrumental techniques have been used for quality control. Particularly, vibrational spectroscopies are a very attractive option, as they can provide high amounts of structural information, leading to the identification of adulterants and simultaneously quantifying them through computational tools such as machine learning. Not to mention important practical advantages, such as allowing for rapid and non-invasive analysis, minimal sample consumption, and compatibility with portable equipment. With this landscape, we have synergically combined the vibrational IR and Raman fingerprints of butter adulterated with margarine to develop data fusion-based models that cover the range from the pure commercial product (butter) to the pure adulterant (margarine). We demonstrate that data fusion models outperform those built with each technique separately, showing that it is the synergy of the vibrational information that improves the final prediction. The predictive ability of our best, final model, evaluated through the Root Mean Square Error of Prediction (RMSEP), was improved by 6.3% compared to the best IR-based model and by 46% compared to the best Raman-based model.