论文部分内容阅读
提出一种结合单类学习器和集成学习优点的Ensembleone-class半监督学习算法.该算法首先为少量有标识数据中的两类数据分别建立两个单类分类器.然后用建立好的两个单类分类器共同对无标识样本进行识别,利用已识别的无标识样本对已建立的两个分类面进行调整、优化.最终被识别出来的无标识数据和有标识数据集合在一起训练一个基分类器,多个基分类器集成在一起对测试样本的测试结果进行投票.在5个UCI数据集上进行实验表明,该算法与tri-training算法相比平均识别精度提高4.5%,与仅采用纯有标识数据的单类分类器相比,平均识别精度提高8.9%.从实验结果可以看出,该算法在解决半监督问题上是有效的.
This paper proposes an Ensemble-class semi-supervised learning algorithm that combines single-class learners and the advantages of integrated learning. The algorithm first establishes two single-class classifiers for two types of data in a small amount of labeled data respectively, and then uses the established two Single-class classifiers jointly identify un-identified samples, and use the identified un-identified samples to adjust and optimize the two established classification surfaces. Finally, the unidentified data and the identified data are finally identified to train a base Classifier and multiple base classifiers are combined to vote on the test results of the test samples.Experiments on five UCI datasets show that the algorithm has an average accuracy of 4.5% higher than the tri-training algorithm, The average recognition accuracy is improved by 8.9% compared with the classifier with purely labeled data.From the experimental results, it can be seen that the algorithm is effective in solving the semi-supervised problem.