基于集成学习算法的慢性肾病早期筛查方法  

An Early Screening Method of Chronic Kidney Disease Based on Ensemble Learning Algorithm

作　　者：姜玉苹 余诚 林燕榕 斯海燕 刘迪 朱江 王浩 陈浩 JIANG Yu-ping;YU Cheng;LIN Yan-rong;SI Hai-yan;LIU Di;ZHU Jiang;WANG Hao;CHEN Hao(ShenTaiWang Healthcare Technology Limited Company,Nanjing 210023,China;National Key Laboratory for Novel Software Technology at Nanjing University,Nanjing University,Nanjing 210023,China)

机构地区：[1]肾泰网健康科技(南京)有限公司,南京210023 [2]南京大学软件新技术国家重点实验室,南京210023

出　　处：《西南大学学报（自然科学版）》2020年第10期17-24,共8页Journal of Southwest University(Natural Science Edition)

基　　金：江苏省重点研发(社会发展)项目(BE20191611);南京市栖霞区发展和改革委员会第二批人工智能企业项目.

摘　　要：慢性肾病是严重危害人类健康的常见疾病,其发病率高,知晓率低.基于集成学习算法的慢性肾病早期筛查方法能够提高肾病知晓率,有利于做到早发现早治疗.搜集2016年到2019年多家医院的体检资料,选取3年内进展为慢性肾病的体检人员作为研究对象,并选取3年内没有进展为慢性肾病的体检人员作为对照组.通过5折交叉验证,采用python 3.7进行随机森林与XGBoost算法模型的训练及测试,通过进展为慢性肾病结局的F1值、真阳性和真阴性指标比较各模型对体检人员3年内是否进展为慢性肾病的预测效果.随机森林算法模型预测效果为,真阳性率0.950,真阴性率0.969,F1值0.957;XGBoost算法模型预测效果为,真阳性率0.966,真阴性率0.955,F1值0.958.Chronic kidney disease,with its high incidence and low awareness,is a common disease that seriously endangers human health.The early screening method of chronic kidney disease based on the ensemble learning algorithm can improve the awareness rate of kidney disease and is conducive to early detection and early treatment.In a study reported herein,the medical examination data of many hospitals from 2016 to 2019 were collected,the examinees who had progressed to chronic kidney disease within three years were selected as the research subjects,and the examinees who had not progressed to chronic kidney disease within three years were taken as the control group.Through 5-fold cross-validation,python 3.7 was used to train and test the random forest and XGBoost algorithm models,and their predictive effect was compared based on the F1-score,and true positive and true negative indicators of the outcome of chronic kidney disease.The prediction effect of the random forest algorithm model was that the true positive rate was 0.950,the true negative rate was 0.969 and the F1-score was 0.957;while that of the XGBoost algorithm model was that the true positive rate was 0.966,the true negative rate was 0.955 and the F1-score was 0.958.

关 键 词：慢性肾病 早期筛查 集成学习 随机森林 XGBoost 交叉验证 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]