基于GMDH的迁移特征选择模型研究*

将迁移学习和数据分组处理算法集成起来,提出了一种基于数据分组处理算法的迁移特征选择(GM-DH-TFS)模型。在UCI的四个数据集上,将GMDH-TFS模型与以全部特征作分类(FULL)的结果以及常用的特征选择模型(前向监督特征选择模型(SFFS)、前向半监督特征选择模型(FW-SemiFS)和迁移特征选择模型(TFS))作比较实验,结果表明,GMDH-TFS在特征选择方面比其他四种方法有更好的效果,在小样本情况下也得到了同样的结果。GMDH-TFS模型可以在数据分布不一致的情况下进行特征选择,同时面对数据匮乏也能取得理想的效果。

Research about transferred feature selection based on GMDH

This paper proposed a transferred feature selection model based on group method of data handling (GMDH-TFS) by integrating transfer learning and the group method of data handling algorithm. Comparison on four data sets of UCI among GMDH-TFS, classification with full features(FULL), supervised forward feature selection (SFFS), forward semi-supervised feature selection(FW-SemiFS) and transferred feature selection(TFS) show that GMDH-TFS has a better performance than other methods as well as in the case of learning with small samples. GMDH-TFS can do feature selection when the data are under different distribution, and can get satisfactory results even the data are not enough.