AI and Raman Spectroscopy Identify Edible Oils in Food.

Ensuring the authenticity of edible oils in processed foods is vital for quality control, fraud prevention, and regulatory compliance. This research presents an integrated framework that combines Raman spectroscopy with various machine learning techniques, including t-SNE, K-means clustering, Decision Trees, and Non-Negative Least Squares (NNLS)-based spectral decomposition. The approach is rooted in Physics-Informed Artificial Intelligence (PI-AI). The study investigated five types of edible oils, both in their pure form and embedded within a fried-potato-chip matrix. Unsupervised analysis revealed clear distinctions between pure oils, but the complex food matrix introduced significant spectral overlap. For pure oils, Decision Trees achieved perfect 100% classification accuracy using only four key Raman variables, representing a mere 0.21% of the original spectral data. When dealing with the more challenging matrix-containing samples, the NNLS-based PI-AI spectral decomposition proved crucial. It effectively separated oil-related spectral signatures from those of the paper and potato, dramatically improving classification. Optimized models achieved accuracies of 86.4% and 85.4% for paper-subtracted and paper-plus-potato-subtracted datasets, respectively, while further reducing the number of important Raman variables to just four or five. This demonstrates that highly accurate, interpretable, and compact spectral representations can be achieved, paving the way for Frugal AI, Edge AI, and portable food-quality monitoring devices.