Open-world quantification of olive-oil adulteration from fluorescence spectra via optimal-transport alignment

Accurate quantification of edible-oil adulteration is constrained by an open-world problem: adulterants encountered at deployment often differ from those seen during training. Here we show that olive-oil volume fraction can be reliably recovered from excitation-emission matrix (EEM) fluorescence spectra even when the adulterant is entirely unseen during training. We propose an optimal-transport (OT) framework that aligns source- and target-domain spectral distributions in an unsupervised manner, requiring no target-domain concentration labels. Under a stringent leave-one-adulterant-out protocol across five adulterant systems, the OT-aligned support vector regression model achieves a mean R 2 of 0.965 (s.d. = 0.004), substantially outperforming raw spectra ( R 2 = 0.831), PCA ( R 2 = 0.693) and PARAFAC ( R 2 = 0.659). These results indicate that olive-oil concentration is encoded as an intrinsic geometric coordinate within the spectral distribution, recoverable by optimal transport without adulterant-specific supervision.