Online hierarchical streaming feature selection based on adaptive neighborhood rough set

Abstract

In the era of open machine learning, a kind of data is accompanied by a hierarchical structure between classes in the label space and the increasing number of features. Therefore, hierarchical classification learning in dynamic changing feature spaces remains an essential research challenge. To address this challenge, we propose an online hierarchical streaming feature selection algorithm based on adaptive neighborhood rough set in this paper, which effectively selects closely interactive features in high-dimensional data with a hierarchical structure. First, a subtree instance set is obtained for each internal node based on the parent–child relationship of the hierarchical structure. Then, an adaptive neighborhood rough set is constructed, and the neighborhood granularity of each instance is auto selected according to the subtree instance set of its parent node. Moreover, an online streaming feature selection framework for hierarchical data is proposed, in which, the streaming feature is evaluated by three steps: online significance analysis, online relevance analysis, and online redundancy analysis. Finally, experiments are conducted on seven hierarchical datasets to evaluate the performance of the proposed algorithm, and extensive results demonstrate that the proposed algorithm outperforms other state-of-the-art online streaming feature selection algorithms.