基于机器学习与SHAP解释的冠心病风险预测模型研究

Research on Coronary Heart Disease Risk Prediction Model Based on Machine Learning and SHAP Interpretation

摘要: 本文基于UCI Cleveland心脏病数据集，采用机器学习方法构建冠心病风险预测模型。对数据进行预处理后，分别建立逻辑回归、支持向量机、随机森林、XGBoost和LightGBM五种分类模型，并结合交叉验证、多指标评价及可靠性分析比较模型性能。同时，利用SHAP方法对最优模型进行解释，分析关键临床特征对预测结果的影响。实验结果表明，五种模型均具有较好的风险识别能力，其中逻辑回归模型综合性能最佳，测试集ROC-AUC达到0.9436。Bootstrap、校准曲线和决策曲线分析结果表明，该模型具有较好的稳定性和应用价值。SHAP分析进一步发现，年龄、最大心率、运动诱发心绞痛和ST段压低等因素对模型预测具有重要影响。基于机器学习结合可解释分析的方法能够有效实现冠心病风险预测，并提高模型结果的透明性，为临床辅助决策提供参考。

Abstract: This paper constructs a risk prediction model for coronary heart disease (CHD) using machine learning methods based on the UCI Cleveland Heart Disease dataset. After data preprocessing, five classification models—Logistic Regression, Support Vector Machine, Random Forest, XGBoost, and LightGBM—were established. Their performances were compared using cross-validation, multi-metric evaluation, and reliability analysis. Meanwhile, the SHAP (SHapley Additive exPlanations) method was applied to interpret the optimal model and to analyze the influence of key clinical features on prediction outcomes. Experimental results show that all five models exhibit good risk discrimination ability, with the Logistic Regression model achieving the best overall performance, attaining a test-set ROC-AUC of 0.9436. Bootstrap resampling, calibration curve, and decision curve analyses indicate that the model has satisfactory stability and clinical utility. SHAP analysis further reveals that age, maximum heart rate, exercise-induced angina, and ST-segment depression are among the most influential factors in model predictions. The proposed approach, combining machine learning with interpretable analysis, can effectively predict CHD risk and enhance the transparency of model outputs, thereby providing a reference for clinical decision support.