【摘要】 心脏病作为全球高致死率及高经济负担的心血管疾病^[1],其精准预测与可解释性分析对临床诊疗至关重要。本研究针对现有预测模型准确性不足及特征解释性弱的问题，提出一种集成特征工程与机器学习的分析框架。首先，通过独立样本T检验（P<0.05）与皮尔逊/斯皮尔曼相关性分析，联合对特征进行预处理分析；然后，利用多种机器学习模型和SHAP^[2-4]方法实现对心脏病的预测及特征重要性排序可视化结果，在UCI心脏病数据集上进行实验，并结合SHAP解释框架实现全局与局部特征重要性量化^[5-9],从而辅助特征选择，进而提供更直观且解释性更强的特征选择策略；最后，挖掘单个特征与心脏病之间的关联，揭示年龄、胆固醇等核心指标的非线性效应及多特征交互机制^[6-10],并且进行合理可解释性分析，为医生对心脏病的精准医疗提供决策支持。更多还原

【Abstract】 Cardiovascular disease is a leading global cause of death with high associated economic burden, making its accurate prediction and interpretable analysis crucial for clinical diagnosis and treatment. Addressing the issues of insufficient accuracy and weak feature interpretability in existing prediction models, this study proposes an analytical framework that integrates feature engineering and machine learning. Firstly, features are pre-processed and analyzed through independent samples t-test(P<0.05) combined with Pearson/Spearman correlation analysis. Subsequently, various machine learning models and the SHAP method are utilized to achieve cardiac disease prediction and visualize feature importance rankings. Experiments are conducted on the UCI heart disease dataset. By combining the SHAP interpretability framework, both global and local feature importance are quantified, thereby assisting feature selection and providing a more intuitive and interpretable feature selection strategy. Finally, the associations between individual features and heart disease are explored, revealing the nonlinear effects of core indicators such as age and cholesterol, along with multi-feature interaction mechanisms. A reasonable interpretability analysis is conducted to provide decision support for physicians in the precise medicine of heart disease.更多还原