流缺失标记环境下的多标记特征选择

在监督学习实际任务中,特征的高维性、标记的动态性和缺失性为监督学习带来严峻的挑战。为解决这些不足,提出流缺失标记环境下的多标记特征选择算法。首先,为解决缺失标记的影响,通过学习标记相关性填补不完整的标记矩阵。其次,利用稀疏学习方法为每个新到达的标记选择类属属性。然后,根据已到达标记的类属属性,通过计算得分选取一个有代表性的特征子集。最后,在11个基准数据集上进行一系列实验表明,所提算法能选择有代表性的特征子集,且分类性能较优。

Multi-label feature selection with streaming and missing labels

In the practical tasks of supervised learning, the high dimensionality of feature space, the dynamic and missing of labels bring severe challenges to supervised learning. To address these problems, a multi-label feature selection with streaming and missing labels algorithm is proposed. Firstly, to solve the impact of missing labels, the missing matrix is completed by learning label correlations. Secondly, sparse learning is utilized to select label-specific features for each newly arrived label. Then, a representative feature subset is selected by calculating the score of each label-specific features of label. Finally, a series of experiments on 11 benchmark data sets demonstrate that the proposed algorithm can effectively select a representative feature subset with better classification performance.