摘要
为提高半监督分类的性能,提出一种基于SOM神经网络的半监督分类算法SSC-SOM。结合SOM的聚类特性,基于先聚类后标记的思想,充分利用有标记样本和未标记样本训练SOM分类器;将聚类的形成和有标记样本分配到各个聚类中同时进行,并根据有标记样本计算各个聚类的聚类中心;在整个未标记样本的范围内,根据聚类中心,使用K近邻算法对未标记样本进行标记,挖掘未标记样本的隐含信息。在UCI数据集中进行分类实验,其结果表明,SSC-SOM的分类率比SSOM提高2.22%,且收敛性较好。
In order to improve the performance of semi-supervised classifier, a kind of semi-supervised classification algorithm SSCSOM is proposed. Based on the clustering characteristics of SOM and the Cluster-then-Label idea, labeled data and unlabeled data are all used to train SOM. The labeled samples are assigned to each cluster and the clusters form simultaneously. The clustering centers are work out. K-NN algorithm is adopted to label the unlabeled samples according to the clustering centers and the information from the unlabeled samples is mined. With UCI dataset, experiments were carried out and the results show that the classification rate of SSC-SOM increases by 2. 22 % than SSOM and the SSC-SOM method had good convergence.
出处
《西华大学学报（自然科学版）》
CAS
2015年第1期36-40,51,共6页
Journal of Xihua University:Natural Science Edition
基金
陕西省教育厅科研计划项目资助(12JK0748)
商洛学院科研项目资助(14SY006
14SKY007)
