摘要
在脑卒中早期风险分类预测领域,现有机器学习算法存在性能不足及临床可解释性匮乏的问题。研究提出一种基于贝叶斯优化算法的XGBoost-SHAP模型,以优化分类预测模型性能并增强其可解释性。该模型应用贝叶斯优化算法对XGBoost模型的超参数进行全局优化,以确定最优参数配置,进而构建高效精准的分类预测模型。实验结果表明,经优化的模型在准确率、F1及AUC值方面分别达到0.930、0.927和0.930。进一步地,研究引入SHAP值对模型进行可解释性分析,量化各特征对预测结果的影响程度。分析结果揭示,年龄与血糖水平存在正相关关系,而吸烟状态与预测结果之间则呈现出复杂的非线性关系。这些发现与临床经验相契合,进一步提升了该模型在临床应用中的透明度和可信度。
In the field of early risk classification and prediction of stroke,existing Machine Learning algorithms have the problems of insufficient performance and lack of clinical interpretability.This study proposes an XGBoost-SHAP model based on Bayesian Optimization(BO)algorithm to optimize the performance of classification prediction models and enhance their interpretability.The model uses BO algorithm to globally optimize the hyperparameters of the XGBoost model to determine the optimal parameter configuration,and then constructs an efficient and accurate classification prediction model.The experimental results show that the values of accuracy,F1 and AUC of the optimized model reach 0.930,0.927 and 0.930,respectively.In addition,this study introduces SHAP values to conduct interpretability analysis of the model to quantify the influence degree for each feature on the prediction results.The analysis results reveal that there is a positive correlation between age and blood glucose level,while there is a complex nonlinear relationship between smoking status and prediction results.These findings are consistent with clinical experience and further enhance the transparency and confidence of the model in clinical application.
作者
王梦豪
史天乐
马云菲
WANG Menghao;SHI Tianle;MA Yunfei(College of Artificial Intelligence,North China University of Science and Technology,Tangshan 063210,China;Hebei Key Laboratory of Industrial Intelligent Perception,Tangshan 063210,China)
出处
《现代信息科技》
2025年第18期17-23,共7页
Modern Information Technology
基金
国家级大学生创新创业训练计划项目(202410081026)
研究生创新资助项目(2025S23)
华北理工大学医工融合专项(XJ2024002801)。
