Combined use of near-infrared spectroscopy, elemental profiles, and stable isotope ratios with machine learning for geographical origin authentication of Sargassum fusiforme

Authentication of the geographical origin of Sargassum fusiforme , a high-value edible seaweed, is crucial for safeguarding its quality and traceability. This study systematically evaluated the discriminative power of near-infrared spectroscopy (NIRS), elemental profiles, and stable isotope ratios, used individually or in combination, for identifying the origin of S. fusiforme from China, Japan, and Korea. Three machine learning algorithms, including linear discriminant analysis (LDA), support vector machine (SVM), and random forest (RF), were compared using 53 samples. Single-feature models showed limited performance, with most prediction accuracies below 70%. Two-way combinations, such as elements combined with NIRS or isotopes combined with elements, improved accuracy and stability, achieving 81.25% with both SVM and RF. The combination of isotopes, elements, and NIRS achieved the highest prediction accuracy of 87.5% with RF. Feature importance analysis identified Cu, As, and δ 15 N as the most influential variables, with importance values of 0.1880, 0.1862, and 0.1751, respectively. These findings suggest that integrating multiple data sources with RF offers a promising strategy for geographical authentication of S. fusiforme .