Abstract: Traditional studies in the identification of organic products mainly focus on detecting certification labels. There were no effective trait indicators for distinguishing organic products. To fill this knowledge gap, we utilized isotope ratio mass spectrometry (IRMS) to differentiate the organic and conventional samples. We selected certified organic farms and adjacent conventional farming areas in Yunnan Province, and collected samples of Polygonatum Sibiricum, lettuce, apples, and soils. Using discriminant analysis (OPLS-DA) and dimensionality reduction analysis, key indicators were identified. Machine learning models such as RandomForest and XGBoost were constructed to develop a method for distinguishing Yunnan’s highland organic products. The results showed that there were significant differences between organic and conventional farming in terms of product δ¹⁵N, soil alkali-hydrolyzable nitrogen and other indicators, with δ¹⁵N being the most important factor for distinguishing organic products. Product δ¹³C, total nitrogen content, soil δ¹⁵N, and alkali-hydrolyzable nitrogen content were significant indicators of organic farming. The machine learning models built using isotope characteristics such as δ¹⁵N and δ¹³C, along with product-soil property parameters, effectively distinguished the organic and conventional products, achieving an accuracy of over 92%.

Key words: stable isotope, organic product, machine learning, classifier, authenticity identification