Soybean (Glycine max (L.) Merril) is a popular foodstuff and crop plant, used in human and animal food. In this work, multielement analysis of soybean grains samples in combination with chemometric tools was used to classify the geographical origins. For this purpose, 120 samples from three provinces of Argentina were analyzed for a panel of 20 trace elements by inductively coupled plasma mass spectrometry. First, we used principal component analysis for exploratory analysis. Then, supervised classification techniques such as support vector machine (SMV) discriminant analysis (SVM‐DA), random forest, k‐nearest neighbors, and class‐modeling techniques such as soft independent modeling of class analogy (SIMCA), potential functions, and one‐class SVM were applied as tools to establish a model of origin of samples. The performance of the techniques was compared using global indexes. Among all the models tested, SVM and SIMCA showed the highest percentages in terms of prediction ability in cross‐validation with average values of 99.3% for SVM‐DA and a median value of balanced accuracy of 96.0%, 91.7%, and 88.3% for the three origins using SIMCA. Results suggested that the developed methodology by chemometric techniques is robust and reliable for the geographical classification of soybean samples from Argentina.

中文翻译：

---

基于微量元素分布的阿根廷生产大豆的可追溯性

大豆 (Glycine max (L.) Merril) 是一种受欢迎的食品和作物植物，用于人类和动物食品。在这项工作中，结合化学计量学工具对大豆籽粒样品进行多元素分析，对地理来源进行分类。为此，通过电感耦合等离子体质谱法分析了来自阿根廷三个省的 120 个样品中的 20 种痕量元素。首先，我们使用主成分分析进行探索性分析。然后，监督分类技术，如支持向量机 (SMV) 判别分析 (SVM-DA)、随机森林、k 最近邻和类建模技术，如类类比的软独立建模 (SIMCA)、潜在函数和一类 SVM 被用作建立样本来源模型的工具。使用全局索引比较了这些技术的性能。在所有测试的模型中，SVM 和 SIMCA 在交叉验证的预测能力方面显示出最高的百分比，SVM-DA 的平均值为 99.3%，平衡准确度的中值分别为 96.0%、91.7% 和 88.3%。使用 SIMCA 的三个起源。结果表明，通过化学计量学技术开发的方法对于阿根廷大豆样品的地理分类是稳健可靠的。