糖尿病患者动脉粥样硬化发病风险预测模型比较研究
A comparative study of risk prediction models for atherosclerosis in patients with diabetes mellitus
韩政元 1杨紫薇 2赵陆洋 1李林怿 3刘奎 3万毅3
作者信息
- 1. 空军军医大学:基础医学院学员三大队,陕西西安 710032
- 2. 空军军医大学:基础医学院学员二大队,陕西西安 710032
- 3. 空军军医大学:卫勤训练基地卫生勤务教研室,陕西西安 710032
- 折叠
摘要
目的 分析比较LightGBM和随机森林机器学习模型在糖尿病患者动脉粥样硬化疾病风险预测模型中的应用.方法 利用国家人口健康科学数据中心公共数据集,使用LightGBM和随机森林两种算法建立动脉粥样硬化疾病风险预测模型并进行比较研究.结果 利用LightGBM和随机森林的机器学习模型,对动脉粥样硬化进行分析发现,随机森林的准确率为0.624 2、曲线下面积(AUC)为0.671 8、精确度为0.629 7,均高于LightGBM,而LightGBM的召回率为0.756 7、F1得分为0.665 2,数值高于随机森林,但二者对于动脉粥样硬化预测效果均较好.结论 在动脉粥样硬化的预测模型中,随机森林的准确率、AUC、精确度更高,LightGBM的召回率、F1得分更高.总体而言,二者均可对动脉粥样硬化进行准确的预测,可以运用到临床实践中,为临床辅助诊断糖尿病并发症的相关研究提供有益借鉴.
Abstract
Objective To analyze and compare the application of LightGBM and random forest machine learning model in the risk prediction model of atherosclerosis in diabetic patients.Methods Based on the public data set of National Population Health Data Center,the risk prediction model of atherosclerosis was established and compared with LightGBM and random forest algorithms.Results The machine learning model of LightGBM and random forest was used to analyze atherosclerosis.It was found that the accuracy of random forest was 0.624 2,area under the curve(AUC)was 0.671 8,and precision was 0.629 7,which were all higher than those of LightGBM.However,the recall rate of LightGBM and F1 score of LightGBM were 0.756 7 and 0.665 2,which were higher than those of random forest,but both of them had good prediction effects for atherosclerosis.Conclusion In the prediction model of atherosclerosis,random forest has a higher accuracy,AUC and precision,while LightGBM has a higher recall rate and F1 score.In general,both of them can accurately predict atherosclerosis,which can be applied to clinical practice and provide useful reference for the related research of clinical auxiliary diagnosis of diabetic complications.
关键词
糖尿病/动脉粥样硬化/机器学习/预测分析Key words
diabetes mellitus/atherosclerosis/machine learning/prediction analysis引用本文复制引用
基金项目
国家自然科学基金面上项目(82073663)
出版年
2024
国家科技期刊平台
NSTL
万方数据