基于半监督学习的单类分类器

半监督学习是一种利用有标记样本和无标记样本进行学习的新的机器学习方法。针对单分类中只有目标类标记样本和大量无标记样本的情况,提出了一种基于半监督学习的单类分类算法。利用已标识的有标记样本建立两个单类分类器,通过相互学习来挖掘未标记样本中的隐含信息,扩大有标记样本的数量。利用所有已标识样本,用不同的单分类方法建立多个单类分类器,通过集成学习的方法得到最终的分类器。在UCI数据集上进行了实验,表明提出的基于半监督学习的单类分类器的有效性。

Semi-supervised learning based on one-class classification and ensemble learning

Semi-supervised learning is a new machine learning method, w hich studies on labeled and unlabeled data. A new semi-superv ised learning algor ithm was propo sed to so lve the one class classification pr oblem w hich has only some labeled posit ive and lot s of unlabeled data. In this pro posed algor ithm tw o one-class classifiers w ere built respectively for the know n labeled targ et data, and then some information under the unlabeled data was mined by the tw o classif iers. And the know n labeled targ et data w ere enlarged by the co-lear ning. The enlarged labeled target data w ere used to t rain the three classifiers, and the ensemble learning w as used to get the f inal classifier. Ex perimental resul ts of the U CI data illust rates that the performance o f the proposed algo rithm is improv ed in some degree compar ed w ith that of the classifier t rained by pur e labeled data