指甲中蕴含的稳定同位素比值信息可用于推断个体的生活地域变化情况，在法庭科学领域个体身份溯源方面具有重要意义。本文采用元素分析-稳定同位素比质谱仪(Elemental Analyzer-Isotope Ratio Mass Spectrometry)对陕西省安康、西安和榆林3个城市常住居民指甲的氢、氧、碳、氮稳定同位素比值进行检测，研究过程采用三种溯源分析模型（判别分析、多层感知器和径向基函数）对所有指甲样本多元稳定同位素比值数据进行分析，并且探讨了不同元素组合对溯源的贡献率。结果表明，三种溯源分析模型均能较好地区分3个城市的指甲样本，多元素组合显著提升了分类准确率，其效果优于单元素、双元素以及三元素分类；其中，判别分析交叉验证正确率为84.7%，多层感知器分类准确率为88.9%，径向基函数模型分类准确率为85.7%。多层感知器模型的分类效果稍优，原因在于其能够有效处理多变量之间的非线性关系。本研究为稳定同位素技术在个体身份识别领域的应用提供了新的研究方向。

Stable isotope ratio information derived from fingernails can be used to infer geographical changes in an individual’s residential area, which holds significant value for personal identification in forensic science. This study utilized Elemental Analyzer-Isotope Ratio Mass Spectrometry (EA-IRMS) to determine the stable isotope ratios of hydrogen (H), oxygen (O), carbon (C), and nitrogen (N) in fingernail samples collected from residents of three cities in Shaanxi Province: Ankang, Xi′an, and Yulin.

These stable isotope ratios act as biological markers that reflect the environmental and geographical conditions experienced by individuals, thereby providing valuable data for tracing an individual’s geographical origin. Three source-tracing analysis models were employed in this research: Discriminant Analysis, Multi-Layer Perceptron (MLP), and Radial Basis Function (RBF). These models were applied to analyze the multivariate stable isotope ratio data obtained from all fingernail samples across the three cities.

Additionally, the study investigated how different combinations of isotopic elements influenced the accuracy of the source-tracing models. The core objective was to evaluate whether the integration of multiple stable isotopes (H, O, C, and N) could enhance the ability to distinguish individuals based on their geographical origins.

The results indicated that all three models effectively differentiated fingernail samples from the three cities, with the MLP model exhibiting the highest accuracy. Compared to single-element or dual-element combinations, the use of multiple isotopes significantly improved classification accuracy. Specifically, the Discriminant Analysis model achieved a cross-validation accuracy of 84.7%, the MLP model reached 88.9%, and the RBF model attained an accuracy of 85.7%. The MLP model performed marginally better, primarily due to its capacity to effectively handle the complex nonlinear relationships between variables in the isotopic data.

These findings underscore the value of integrating multiple isotopic elements for more precise source tracing and highlight the potential of isotope analysis in accurate geographical identification. The research demonstrated that combining different stable isotope elements—particularly H, O, C, and N—provides a more comprehensive and reliable approach to tracing an individual’s origin.

Furthermore, the results emphasize the importance of advanced statistical models in forensic analysis, especially for identifying individuals using biological samples such as fingernails. This study offers valuable insights into the application of stable isotope technology in personal identification and opens new avenues for forensic science applications, particularly in geographical origin tracing. Ultimately, this research contributes to enhancing the accuracy of forensic investigations and personal identification processes in both criminal and civil contexts.