三重集约束下的自适应SSLBoost分类方法  

Adaptive SSLBoost classification algorithm based on triple set constraints

作　　者：原鑫 王振友[1] Yuan Xin;Wang Zhenyou(School of Applied Mathematics,Guangdong University of Technology,Guangzhou 510520,China)

机构地区：[1]广东工业大学应用数学学院,广州510520

出　　处：《计算机应用研究》2020年第8期2376-2380,共5页Application Research of Computers

基　　金：广州市科技计划资助项目(201707010435)。

摘　　要：在分类问题中,常用的高效算法有半监督学习算法、Bagging算法和Boosting算法等,当标记数据很少、数据间差异较大时,很难找到有效的规则来分类。针对此问题提出了三重集约束下的Boosting分类算法,对标记数据、伪标记数据、无标记数据进行三重约束划分;同时引入平衡函数将更新数据的近邻两点加权,确立数据空间稳定点;根据稳定点信息对分类器进行迭代,采用梯度下降法使得平衡函数收敛,得到最终的伪标记数据和分类器。经过UCI九个数据集的实验,验证了该算法更为高效、可行。In the classification problem,the commonly algorithms are semi-supervised learning algorithm,Bagging and Boosting classification algorithm.When there is little mark data and the difference between data is large,it’s hard to find effective regulations to classify them.This paper proposed a Boosting classification algorithm under triple set constraints to solve this problem.The algorithm divided the mark data,pseudo mark data and unmarked data by triple constraints,and introduced a balance function to weight the two neighbors of the mark data and established the stable data for this space.Then it iterated the classifier according to that data information.Through the gradient descent method,the balance function converged and finally got the pseudo-marker data and classifier.Experiments on nine datasets of UCI verify that this algorithm is more efficient and feasible.

关 键 词：BOOSTING算法 空间稳定点 三重集约束 梯度下降 平衡函数 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]