粗糙互信息的不平衡多标记特征选择算法

Feature Selection Algorithm with Imbalanced Multi-labels Based on Rough Mutual Information

特征选择作为处理多标记学习中数据高维性的一种有效方法,得到了众多学者的研究与关注。由于部分特征仅仅与某些标记有着强相关性而与整个标记空间的相关性不强,不能简单通过与标记空间整体的相关性判断取舍。此外,多标记的分布是不平衡的。因此,根据标记密度对标记空间进行划分,并分别进行相关性的判断,同时在不同标记空间进行不同比例的采样。引入具有补的性质的粗糙熵代替传统熵的度量方式,提出了基于粗糙互信息的不平衡多标记特征选择算法,在5个公开数据集上的实验结果表明了算法的有效性。

Feature selection,as an effective method to deal with the high dimensionality of data in multi-label learning,has attracted attention wiclely.Since some features only have strong correlations with certain labels but not with the entire label space,it is not possible to simply make a decision through the correlation.In addition,the distribution of multi-label is imbalanced,so the label space is divided according to the density of the labels and the correlation is judged separately.Meanwhile,different proportions of sampling are performed in different label spaces.The rough entropy with complementary properties is also introduced instead of the traditional entropy measurement method,and the feature selection algorithm with imbalanced multi-labels based on rough mutual information is proposed.Experimental results on five public data sets show the effectiveness of the algorithm.