Tensor-based unsupervised feature selection for error-robust handling of unbalanced incomplete multi-view data

 Recent advancements in multi-view unsupervised feature selection (MUFS)  have been notable, yet two primary challenges persist. First, real-world  datasets frequently consist of unbalanced incomplete multi-view data, a  scenario not adequately addressed by current MUFS methodologies.  Second, the inherent complexity and heterogeneity of multi-view data  often introduce significant noise, an aspect largely neglected by  existing approaches, compromising their noise robustness. To tackle  these issues, this paper introduces a Tensor-Based Error Robust  Unbalanced Incomplete Multi-view Unsupervised Feature Selection  (TERUIMUFS) strategy. The proposed MUFS framework specifically caters to  unbalanced incomplete multi-view data, incorporating  self-representation learning with a tensor low-rank constraint and  sample diversity learning. This approach not only mitigates errors in  the self-representation process but also corrects errors in the  self-representation tensor, significantly enhancing the model’s  resilience to noise. Furthermore, graph learning serves as a pivotal  link between MUFS and self-representation learning. An innovative  iterative optimization algorithm is developed for TERUIMUFS, complete  with a thorough analysis of its convergence and computational  complexity. Experimental results demonstrate TERUIMUFS’s effectiveness  and competitiveness in addressing unbalanced incomplete multi-view  unsupervised feature selection (UIMUFS), marking a significant  advancement in the field.