中国儿童保健杂志 ›› 2025, Vol. 33 ›› Issue (10): 1114-1120.DOI: 10.11852/zgetbjzz2025-0336

• 儿童代谢性疾病专栏 • 上一篇    下一篇

基于生物电阻抗与决策树模型的学龄前儿童超重肥胖筛查与亚型识别研究

马瑞, 沈张逸霏, 陆乐, 唐锦阳   

  1. 上海师范大学体育学院,上海 201400
  • 收稿日期:2025-04-14 修回日期:2025-07-09 发布日期:2025-10-11
  • 通讯作者: 沈张逸霏,E-mail:1634530372@qq.com
  • 作者简介:马瑞(1973—),女,博士,主要研究方向为学校体育、幼儿体育。
  • 基金资助:
    上海市教育科学研究项目(C2022113)

Screening and subtype identification of overweight and obesity in preschool children based on bioelectrical impedance analysis and decision tree modeling

MA Rui, SHEN Zhangyifei, LU Le, TANG Jinyang   

  1. College of Physical Education, Shanghai Normal University, Shanghai 201400, China
  • Received:2025-04-14 Revised:2025-07-09 Published:2025-10-11
  • Contact: SHEN Zhangyifei, E-mail: 1634530372@qq.com

摘要: 目的 基于生物电阻抗技术(BIA)与决策树模型,构建肥胖亚型分类规则,旨在识别学龄前儿童超重肥胖的敏感指标及其亚型特征,为精准干预提供依据。方法 于2024年3—12月采用方便抽样选取上海市550名3~6岁学龄前儿童,通过INBODY J30测定身体成分指标,对比BMI与体脂百分比(PBF)的肥胖检出差异;通过共线性分析筛选出关键变量,采用CRT决策树构建亚型分类规则,并借助XGboost评估特征重要性,通过10折交叉验证模型的稳健性。结果 BMI标准超重肥胖检出率为26.5%,而PBF标准为21.78%。体成分分析显示,超重肥胖儿童与正常体重儿童有29项指标存在显著差异,通过逐步回归分析解决变量间的共线性问题,保留10项变量进入决策树模型构建。基于CRT算法生成的决策树包含17个节点(9个终端节点),将超重肥胖分为三型:下肢脂肪主导型(节点10)、骨骼肌质量低下型(节点16)和高肌肉量型(节点11),模型总体准确率达97.5%。结论 基于BIA与决策树的多维筛查模型可精准识别学龄前儿童超重肥胖亚型,研究提出的亚型分类框架为制定差异化干预策略提供参考和依据,对优化儿童肥胖早期防控措施具有重要实践价值。

关键词: 决策树, 学龄前儿童, 超重, 肥胖, 预测模型

Abstract: Objective To develop a classification framework for overweight/obesity subtypes based on bioelectrical impedance analysis (BIA) and decision tree modeling, with the goal of identifying sensitive indicators and characteristic features of overweight/obesity subtypes among preschool children, so as to provide a basis for targeted interventions. Methods A convenience sample of 550 preschool children aged 3 - 6 years was recruited in Shanghai from March to December 2024.Body composition was measured using the INBODY J30 device.Differences in obesity detection between body mass index (BMI) and percent body fat (PBF) were compared.Key variables were selected through collinearity analysis.A classification and regression tree (CRT) algorithm was applied to construct subtype classification rules.Feature importance was evaluated using XGBoost, and model robustness was assessed via 10-fold cross-validation. Results The detection rate of overweight/obesity was 26.5% based on BMI criteria and 21.78% based on PBF criteria.Body composition analysis revealed 29 indicators that significantly differed between overweight/obese and normal-weight children.Stepwise regression was used to address multicollinearity, resulting in 10 variables being included in the decision tree model.The CRT-generated tree contained 17 nodes (9 terminal nodes) and classified obesity into three subtypes: lower-body fat-dominant (Node 10), low skeletal muscle mass (Node 16), and high muscle mass (Node 11).The overall accuracy of the model reached 97.5%. Conclusions A multidimensional screening model integrating BIA and decision tree analysis can accurately identify overweight/obesity subtypes in preschool children.The proposed subtype classification framework offers valuable insights for developing differentiated intervention strategies and has practical significance for optimizing early prevention and control measures for childhood obesity.

Key words: decision tree, preschool children, overweight, obesity, prediction model

中图分类号: