Abstract: Objective To identify and screen characteristic volatile components in human hand and foot odors associated with gender and age, thereby characterizing the attributes of different population groups. Methods Thermal desorption-gas chromatography-mass spectrometry (TD-GC-MS) was used to analyze the volatile compounds from human hands and feet. Single-factor and multi-factor analyses were used to identify different components related to gender and age across various populations. Genetic algorithm-random forest (GA-RF) machine learning techniques were used to predict the characteristic components of different genders and ages, and to construct a classification model. Results A total of 304 volatile components were identified from human hand and foot odors. They were discerned through t-tests and orthogonal partial least squares-discriminant analysis (OPLS-DA). Components with P<0.05 and VIP>1 were selected. Genetic algorithm (GA) was used to optimize random forest (RF) to construct a judgment model. The accuracies of gender identification using hand and foot features were 92.02% and 81.46%, respectively, and those for age identification were 76.13% and 73.49%, respectively. Conclusion Based on statistics and GA-RF machine learning methods, gender and age difference markers in human hand and foot odor were screened, and a prediction model was constructed, providing a novel approach for the biometric characterization of human odor in forensic science.

Key words: thermal desorption, gas chromatography-mass spectrometry (GC-MS), hand and foot odor, genetic algorithm (GA), random forest (RF), identification

摘要： 目的 筛选人体手足气味中与性别、年龄相关的特征组分，进行不同人群性别与年龄特征刻画。方法 采用热解吸-气相色谱-质谱法（thermal desorption-gas chromatography-mass spectrometry，TD-GC-MS）检测人体手足中的挥发性气味信息，利用单因素分析与多因素分析筛选出不同性别、年龄人群手足气味中的差异组分，并通过遗传算法-随机森林 （genetic algorithm-random forest，GA-RF）机器学习方法预测不同性别、年龄的特征组分，并构建判别模型。结果 从人体手足部位中共检测出 304 种挥发性物质，通过t检验和正交偏最小二乘判别分析（orthogonal partial least squares-discriminant analysis, OPLS-DA）筛选P<0.05且变量投影重要性（variable importance in projection，VIP）>1的差异组分，使用遗传算法（genetic algorithm，GA）优化随机森林（random forest，RF）算法构建判别模型，利用手足特征进行性别识别的准确率分别为92.02%和81.46%，年龄识别的准确率分别为76.13%和73.49%。结论 基于统计学和GA-RF机器学习方法，筛选出人体手足气味中不同性别、年龄的差异标志物，构建判别模型，为人体气味在法庭科学领域中的应用提供新思路。

关键词: 热解吸, 气相色谱-质谱法, 手足气味特征, 遗传算法, 随机森林, 识别