基于集成学习白蛋白水平预测模型的可解释性研究OA
Research on the interpretability of an ensemble learning-based model for predicting albumin levels
白蛋白水平是反映患者营养风险的重要指标.针对白蛋白水平预测数据集缺乏、数据冗余、研究缺乏可解释性的问题,构建完整的营养支持数据集,提出一种集成学习的框架,用于老年患者的白蛋白水平预测.在该研究中,利用基于树结构的贝叶斯优化(TPE)算法优化后的XGBoost(TPE-XGBoost)模型,与其他4种基线模型比较证明其性能,结合模型解释算法(SHAP)进行特征选择并从全局和局部的层面增强模型的可解释性.结果表明,TPE-XGBoost模型简单有效,预测准确率达87.8%,此外,SHAP进行可解释分析得到4项关键因素以及2项因素的阈值效应与临床研究可以互相验证.
Albumin level is an important indicator of nutritional risk of patients.Aiming at the prob-lems of the lack of albumin level prediction dataset,data redundancy and lack of interpretability of studies,a complete nutrition support dataset was constructed and an integrated learning framework was proposed for albumin level prediction in elderly patients.In this study,the XGBoost model opti-mized by Tree-structured Parzen Estimator(TPE)algorithm(TPE-XGBoost)was compared with other 4 baseline models to prove its performance,and the model interpretation algorithm SHapley Ad-ditive exPlanations(SHAP)was combined to select features and enhance the interpretability of the model from the global and local levels.The results showed that the TPE-XGBoost model was simple and effective,and the prediction accuracy reached 87.8%.In addition,the interpretable analysis of SHAP obtained 4 key factors and the threshold effects of 2 factors could be mutually verified with clinical studies.
督静雯;滕飞;林宁;李运明
西南交通大学 计算机与人工智能学院,四川 成都 611756||西部战区总医院 营养科,四川 成都 610083西南交通大学 计算机与人工智能学院,四川 成都 611756西部战区总医院 营养科,四川 成都 610083西部战区总医院 医疗保障中心信息科,四川 成都 610083
医药卫生
白蛋白水平预测模型可解释性XGBoost模型解释算法树结构的贝叶斯优化集成学习
albumin level predictionmodel interpretabilityXGBoostSHAPTPEensemble learning
《常州大学学报（自然科学版）》 2026 (3)
82-92,11
四川省干部保健科研课题资助项目(川干研2051-1303).
评论