一种协同半监督分类算法Co-S3OM

为了提高半监督分类的有效性, 提出了一种基于SOM神经网络和协同训练的半监督分类算法Co-S3OM (coordination semi-supervised SOM)。将有限的有标记样本分为无重复的三个均等的训练集, 分别使用改进的监督SSOM算法(supervised SOM)训练三个单分类器, 通过三个单分类器共同投票的方法挖掘未标记样本中的隐含信息, 扩大有标记样本的数量, 依次扩充单分类器训练集, 生成最终的分类器。最后选取UCI数据集进行实验, 结果表明Co-S3OM具有较高的标记率和分类率。

One of semi-supervised classification algorithm named Co-S3OM based on cooperative training

In order to improve the classification effectiveness of semi-supervised classification, this paper designed a semi-supervised classification algorithm based on SOM neural network and collaborative training. It divided the limited labeled samples into three equal training sets and respectively used supervised SOM algorithm to train the three single classifier. It mined the implicit information of unlabeled samples by voting together through three single classifier to expand the number of labeled samples, extend a single classifier training set and generate the final classifier. Finally, experiment results conduct on UCI data set show that Co-S3OM has a higher labeling rate and classification rate.