基于贝叶斯优化的可解释XGBoost脑卒中风险分类研究

打开文本图片集

中图分类号：TP301.6 文献标识码：A 文章编号：2096-4706（2025）18-0017-07

Abstract：Inthefeldofearlyrisk classificationand predictionof stroke，existing Machine Learningalgorithms have the problemsofinsuffcient performance andlackofclinicalinterpretability.Thisstudyproposes an XGBoost-SHAPmodelbased onBayesian Optimization （BO）algorithm tooptimize the performanceofclasification predictionmodelsand enhance their interpretability.The modeluses BOalgorithm togloballoptimize thehyperparameters ofthe XGBoost model todetermine the optimal parameterconfiguration，andthenconstructsanefcientandaccurateclasification predictionmodel.Theexperimental results show thatthevaluesofaccuracy，F1andAUCoftheoptimized model reach0.930，0.927and0.930，respectively.In addition，thisstudyintroduces SHAPvalues toconductinterpretabilityanalysisofthe model toquantifytheinfluencedegree foreach featureonthepredictionresults.Theanalysisresultsreveal thatthereisapositivecorrelationbetweenageand blood glucose level， whilethereisacomplex nonlinearrelationship betweensmoking status and predictionresults.Thesefindings are consistent withclinical experienceandfurther enhancethe transparencyandconfidenceof themodelinclinicalapplication.

Keywords： stroke risk; XGBoost; explainable algorithm; Bayesian Optimization algorithm

0 引言

脑卒中是一种由脑部血液循环障碍引发的急性脑血管疾病，具有高发病率、高致死率和高致残率的特点，严重威胁着人类生命健康[1。（剩余8841字）