基于均衡分类的脑卒中风险预测模型  

Risk prediction model of stroke based on balance classification

作　　者：曹艳[1] 殷旭[1] 

机构地区：[1]北京信息科技大学计算机学院,北京100085

出　　处：《计算机工程与设计》2015年第9期2508-2513,共6页Computer Engineering and Design

基　　金："十二五"国家科技支撑计划基金项目(2012BAJ18B07-05);北京市人才培养模式创新试验区基金项目(京教函[2009]630号);北京市教委科技面上基金项目(KM201511232012)

摘　　要：为解决不均衡脑卒中数据分类偏倚问题,提出一种基于均衡分类的脑卒中风险预测模型,其主要算法是基于不均衡数据的分类算法SMOTE&PAM-means+C4.5。使用智能型过抽样技术SMOTE生成新的少数类数据;使用聚类算法PAM-means对多数类数据进行聚类,将相似样本聚到一簇后,对各簇按比例进行抽样,形成新的多数类数据;使用C4.5算法对新的多数类和少数类数据形成的均衡数据集进行分类。实验结果表明,该模型可以有效解决分类结果偏倚问题,降低少数类数据分类误差,提高分类准确度,为脑卒中风险的准确预测提供依据。To solve classification bias problem of imbalanced data of stroke, a stroke risk prediction model based on balanced classification was proposed, whose main algorithm is a classification algorithm based on imbalanced data, namely SMOTE＆PAM-means＋CA. 5. SMOTE technology （synthetic minority oversampling technique） was used to generate new mi- nority class data Clustering algorithm PAM-means was used to cluster majority class data to make similar samples in a class for sampling each class proportionally. CA. 5 algorithm was used to classify the balanced data set generated using the new minority and majority data. Experimental results show that the model can effectively avoid the classification bias problem, reduce classification errors of minority class and improve the classification accuracy to provide more effective supports for more accurate prediction of stroke risk.

关 键 词：C4.5算法 SMOTE算法 K-MEANS聚类算法 PAM算法 不均衡数据 偏倚问题 脑卒中 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]