摘要
利用机器学习算法,改变传统心血管疾病(CVD)预测模型的严格数理化公式,以增加危险因素的纳入、降低数据格式的要求.首先提出利用基于单隐层前馈神经网络(SLFNs)的极限学习机(ELM)算法建立CVD预测模型;进一步通过五阶段连续变异方式建立增强领导粒子的粒子群算法(ELPSO),以粒子群(PSO)算法的优化策略,对SLFNs的隐层单元参数进行优化.通过对UCI数据库Statlog (heart)数据集和heart disease database分析结果显示,所提ELPSO-ELM模型测试正确率分别达到85. 71%、84. 00%,AUC(ROC曲线下面积)分别达到0. 902 4、0. 842 3,高于传统CVD预测模型,同时放松了数据线性化约束,能纳入更多的复杂危险因素.
In order to increase the risk factors that could be accepted and reduce the data format requirements in cardiovascular disease(CVD)prediction models,machine learning algorithms were used to change the strict mathematical formulas of traditional CVD prediction models.Firstly,a CVD prediction model by extreme learning machine(ELM)algorithm based on single hidden layer feed-forward neural network(SLFNs)was proposed.Further more,an enhanced leader particle swarm optimization(ELPSO)through a five-staged successive mutation method was used,and the optimized strategy of PSO was also used to optimize the SLFNs hidden layer units parameters.The analysis results on Statlog(Heart)dataset and Heart Disease Dataset of UCI database indicated that the test accuracy of proposed ELPSO-ELM model could reach 85.71%and 84.00%respectively,the AUC(The area under the ROC curve)could reach 0.902 4 and 0.842 3 respectively.They were higher than conventional CVD prediction models.The proposed model relaxed the linear constraints of data format and more complex risk factors could be accepted.
作者
逯鹏
李奇航
尚莉伽
李新建
张微
LU Peng;LI Qihang;SHANG Lijia;LI Xinjian;ZHANG Wei(School of Electrical Engineering,Zhengzhou University,Zhengzhou 450001,China;Collaborative Innovation Center of Internet Medical and Healthcare in Henan,Zhengzhou 450001,China;Primary and Secondary School Health Care in Beijing Dongcheng District,Beijing 100007,China)
出处
《郑州大学学报（工学版）》
CAS
北大核心
2019年第2期1-5,共5页
Journal of Zhengzhou University（Engineering Science）
基金
国家自然科学基金资助项目(60841004
60971110
61172152)
郑州市科技攻关资助项目(112PPTGY219-8)
河南省青年骨干教师资助计划(2012GGJS-005)
关键词
心血管疾病
风险预测
极限学习机
粒子群
cardiovascular disease
risk assessment
extreme learning machine
particle swarm optimization
