An incomplete multiview clustering approach considering missing data recovery based on consistency.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Real-world multiview data frequently exhibits complex missingness patterns, significantly degrading clustering performance.
• Existing methods often neglect inter-view consistency or provide unreliable missing data recovery.
• Dimensional heterogeneity across views poses a challenge for effective clustering.

Purpose of the Study:

• To propose an incomplete multiview clustering algorithm for reliable missing data recovery.
• To enhance clustering performance by preserving inter-view consistency.
• To address dimensional heterogeneity and improve data completion through a novel approach.

Main Methods:

• Constructing a shared latent subspace representation across views.
• Employing adaptive graph learning to align local view structures with a global consensus graph.
• Utilizing clustering metrics of non-missing samples to guide iterative optimization of missing data.
• Implementing a view weight assignment strategy based on consensus graph differences.

Main Results:

• The proposed method achieves reliable recovery of missing data.
• Clustering optimization is achieved synchronously with data complementation.
• Experimental results demonstrate superior performance compared to existing approaches on multiple datasets.

Conclusions:

• The developed algorithm effectively handles incomplete multiview data by integrating data recovery and clustering.
• Preserving inter-view consistency is crucial for robust performance in multiview clustering.
• The adaptive graph learning and view weighting strategies contribute to improved clustering accuracy.