.png){kind=icon}
.png){kind=icon}
Return
Sequential subspace clustering via joint feature selection and spatial-temporal graph learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.131280.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Conventional subspace clustering constructs representations directly from raw features, which make it highly susceptible to redundant or irrelevant dimensions. This drawback becomes even more pronounced in sequential data, where meaningful relationships are further governed by complex spatial-temporal dependencies that are rarely modeled explicitly. To jointly address these challenges, we propose a novel sequential subspace clustering framework that jointly integrates feature selection and spatial-temporal graph learning (JFS2GL). In this framework, feature selection is first employed to identify the most informative dimensions for each sample, which are then used to construct spatial weights. A spatial-temporal graph with ℓ1-norm regularization is subsequently designed to capture intrinsic spatial-temporal relationships in the sequence. Furthermore, by exploiting the linear dependency among neighboring subspaces in sequential data, we design a weighted spatial-temporal constraint that reinforces the block-diagonal structure of the representation matrix. The final model integrates low-rank representation, the ℓ1-regularized spatial-temporal graph, and the proposed weighted constraint. To efficiently solve the resulting optimization problem, we develop a linearized alternating direction method. Extensive experiments on real-world datasets demonstrate the effectiveness and competitiveness of JFS2GL compared to state-of-the-art methods.
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W