Multi-label Feature Selection with Adaptive Subspace Learning

In recent years, with the development of the internet, large-scale multi-label datasets have continuously emerged in various domains. However, these datasets often contain numerous redundant or irrelevant features, significantly impacting downstream tasks in multi-label learning. Therefore, multi-label feature selection, an effective dimensionality reduction method, has attracted the attention of many researchers. Nevertheless, existing multi-label feature selection algorithms struggle to effectively consider label correlations, which are crucial in multi-label learning, leading to decreased performance. In this paper, we propose a multi-label feature selection algorithm based on adaptive subspace learning, named “Adaptive Subspace Feature Selection” (ASFS), to address this issue. Specifically, we first partition the label space into a low-dimensional subspace to learn label correlations. Subsequently, we achieve feature selection based on the mapping relationship between the original feature space and the subspace. Furthermore, we enhance the accuracy of label correlations by maintaining consistency between the subspace and the feature space, achieving adaptive subspace learning. The effectiveness and robustness of the proposed algorithm are validated through extensive experiments.