基于改进Tri-Training算法的健康大数据分类模型研究  

Research on Healthy Big Data Classification Model Based on Improved Tri-Training Algorithm

作　　者：唐校辉 廖欣[3] 陈雷霆[1,4] 陈文昭[5] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都611731 [2]东莞成电金盘软件技术有限公司,东莞523808 [3]四川大学华西妇产儿童医院,成都611731 [4]电子科技大学广东电子信息工程研究院,东莞523808 [5]东莞东华医院,东莞523808

出　　处：《现代计算机（中旬刊）》2017年第7期21-25,共5页Modern Computer

基　　金：广东省省级科技计划项目(No.2014A090906004)

摘　　要：Tri-Training是半监督协同训练的代表性算法之一,它运用统计技术标记置信度,并结合噪音学习理论进行无标记样本分类。当扩充样本训练集不满足噪音学习理论时,会进行随机采样,针对传统Tri-Training算法随机选取基础分类器的扩充训练样本集会引入噪声这一缺陷,通过更改扩充样本训练集选取方式,剔除可能提高分类误差的样本。在健康大数据集上进行一系列验证试验,实验结果表明,改进的算法优于原始算法,降低分类错误率。Tri-Training is a represented algorithm for semi-supervised co-training, it uses statistical techniques to mark the confidence and combine the noise learning theory to classify the unmarked sample. When the extended sample training set does not satisfy the noise learning theory, random sampling is performed. In order to solve the disadvantages of the traditional Tri-Training algorithm which may introduce noise when select the extended sample training set, changes the method of selecting the extended training set, and removes the sample which would improve the classification errors. Carries out a series of verification experiments on the health data set. The experimental results show that the improved algorithm is superior to the original algorithm and reduces the error probability of classification.

关 键 词：Tri—Training 协同训练 分类误差 噪声样本 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]