“三减三健”政策下基于XGBoost与SHAP分析的可解释性肥胖风险预测研究

An Explanatory Study on Obesity Risk Prediction Based on XGBoost and SHAP Analysis under the “Three Reductions and Three Healthy Conditions” Policy

摘要: 在全球肥胖问题日益严峻及中国“三减三健”政策推进的背景下，本研究基于墨西哥、秘鲁和哥伦比亚三国2111份多维度健康数据集，提出一种基于XGBoost与SHAP分析的可解释性肥胖风险预测方法。首先基于XGBoost集成算法构建肥胖风险预测模型(模型准确率达96.22%)，并根据XGBoost内嵌评估指标对模型进行初步特征解释。运用Tree SHAP方法对诊断模型进行特征解释分析，探究重要特征对肥胖风险的影响关系，分析特征之间的依赖交互效应，直观、透明地揭示模型的预测机制。最后利用聚类分析将人群划分为6类，结合中国“体重管理年”政策提出精准干预策略。研究验证了“饮食–运动”干预路径的跨文化有效性，为《健康中国2030》的全球化经验输出提供循证依据，开创了健康政策国际互鉴的新范式。

Abstract: Against the background of the increasingly serious global obesity problem and the promotion of China’s “Three Reductions and Three Healthy Conditions” policy, this study proposes an interpretable obesity risk prediction method based on XGBoost and SHAP analysis based on 2111 multidimensional health datasets from Mexico, Peru and Colombia. An obesity risk prediction model was first constructed based on the XGBoost integration algorithm (with a model accuracy of 96.22%), and the initial features of the model were interpreted based on the XGBoost embedded assessment metrics. The Tree SHAP method was used to analyze the feature interpretation of the diagnostic model, to explore the relationship of important features on obesity risk, to analyze the dependency interaction effect between features, and to reveal the prediction mechanism of the model intuitively and transparently. Finally, cluster analysis was used to classify the population into 6 categories, and precise intervention strategies were proposed in accordance with China’s “Year of Weight Management” policy. The study validates the cross-cultural effectiveness of the “diet-exercise” intervention pathway, provides evidence-based basis for the globalization of the “Healthy China 2030”, and creates a new paradigm for the international mutual understanding of health policies.