Partition-Level Tensor Learning-Based Multiview Unsupervised Feature Selection.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• Multiview unsupervised feature selection is crucial for dimensionality reduction and pattern identification in complex datasets.
• Existing methods often neglect view-specific performance improvements, discriminative partition information, and the impact of marginal samples.

Purpose of the Study:

• To propose a novel method, partition-level tensor learning-based multiview unsupervised feature selection (PTFS), addressing limitations of current approaches.
• To enhance feature selection by integrating high-order view correlations and discriminative partition information.
• To mitigate the negative effects of marginal samples in multiview data.

Main Methods:

• PTFS optimizes a low-rank constrained tensor derived from base partition matrices.
• A statistic-based adaptive self-paced strategy prioritizes confident samples for model training.
• An alternating optimization approach is employed to solve the proposed model.

Main Results:

• Extensive experiments on ten datasets validate the effectiveness of PTFS.
• The proposed method demonstrates superior performance compared to state-of-the-art techniques.
• PTFS shows efficiency in multiview unsupervised feature selection.

Conclusions:

• PTFS effectively addresses key limitations in multiview unsupervised feature selection.
• The method successfully leverages complex data structures for improved feature selection.
• The proposed approach offers a significant advancement in the field of machine learning.