Scholarlink

Background <p>Despite the well-documented benefits of physical activity, insufficient activity remains highly prevalent among adolescents with obesity. This study is the first to apply interpretable machine learning methods to identify the barriers, facilitators, and U-shaped determinants of physical activity in this population.</p> Methods <p>We analyzed data from 1,041 adolescents with obesity from the China Education Panel Survey. A range of personal, family, and school-level variables were incorporated to construct six machine learning models for predicting physical activity attainment. The Shapley Additive Explanations (SHAP) method, a game-theoretic approach for explainable artificial intelligence, was used to identify key predictive factors and quantify their relative contributions.</p> Results <p>The Random Forest model demonstrated the best performance, achieving an accuracy of 85.30% and an AUC of 0.720 on the test set. SHAP analysis revealed several key factors associated with physical activity. Positive facilitators included parents with an education level beyond high school (≥ 3.27 for mothers, ≥ 3.51 for fathers), higher school rankings (≥ 3.77), adequate school sports facilities (≥ 1.44), and a personal interest in sports. Negative barriers included excessive screen time (≥ 5.51&#xa0;h) and school location in central urban areas (≥ 4.23). Notably, U-shaped relationships were identified for academic workload, sleep problems, and self-perceived appearance. Specifically, moderate levels of these factors were associated with lower physical activity, whereas both low and high extremes promoted activity.</p> Conclusion <p>This study demonstrates that physical activity among adolescents with obesity is shaped by a complex interplay of individual, family, and school-level factors, with parental education emerging as the strongest predictor. The identification of specific risk thresholds (e.g., screen time ≥ 5.51&#xa0;h) and U-shaped relationships offers precise, actionable targets for intervention. These findings underscore the need for multi-level strategies: families should prioritize fostering cultural capital, schools should ensure facility accessibility beyond regular hours, and policymakers must address environmental constraints in urban settings. To facilitate the translation of these insights into practice, a web-based tool has been deployed to help physical education teachers identify at-risk students and design targeted interventions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Why they do not move: An explainable machine learning analysis of physical activity barriers in obese adolescents and tool translation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月1日

查看全文
  • Cheng Chen,
  • Kai Chen,
  • Wenqian Du,
  • Shengtao Li,
  • Wenling Gou,
  • Jing Yang,
  • Pedro Forte,
  • Xiaoran Zhang,
  • Yuwen Shangguan,
  • Yongyu Huang,
  • Hao Zhang,
  • Xiaofei Zhang,
  • Zhiyi Lin,
  • Xiaolin Yao,
  • Huan Li

摘要

Background

Despite the well-documented benefits of physical activity, insufficient activity remains highly prevalent among adolescents with obesity. This study is the first to apply interpretable machine learning methods to identify the barriers, facilitators, and U-shaped determinants of physical activity in this population.

Methods

We analyzed data from 1,041 adolescents with obesity from the China Education Panel Survey. A range of personal, family, and school-level variables were incorporated to construct six machine learning models for predicting physical activity attainment. The Shapley Additive Explanations (SHAP) method, a game-theoretic approach for explainable artificial intelligence, was used to identify key predictive factors and quantify their relative contributions.

Results

The Random Forest model demonstrated the best performance, achieving an accuracy of 85.30% and an AUC of 0.720 on the test set. SHAP analysis revealed several key factors associated with physical activity. Positive facilitators included parents with an education level beyond high school (≥ 3.27 for mothers, ≥ 3.51 for fathers), higher school rankings (≥ 3.77), adequate school sports facilities (≥ 1.44), and a personal interest in sports. Negative barriers included excessive screen time (≥ 5.51 h) and school location in central urban areas (≥ 4.23). Notably, U-shaped relationships were identified for academic workload, sleep problems, and self-perceived appearance. Specifically, moderate levels of these factors were associated with lower physical activity, whereas both low and high extremes promoted activity.

Conclusion

This study demonstrates that physical activity among adolescents with obesity is shaped by a complex interplay of individual, family, and school-level factors, with parental education emerging as the strongest predictor. The identification of specific risk thresholds (e.g., screen time ≥ 5.51 h) and U-shaped relationships offers precise, actionable targets for intervention. These findings underscore the need for multi-level strategies: families should prioritize fostering cultural capital, schools should ensure facility accessibility beyond regular hours, and policymakers must address environmental constraints in urban settings. To facilitate the translation of these insights into practice, a web-based tool has been deployed to help physical education teachers identify at-risk students and design targeted interventions.