A Multi-Kernel-Based Multi-View Deep Non-Negative Matrix Factorization for Enhanced Healthcare Data Clustering

• Published in: IEEE transactions on consumer electronics Vol. 71; no. 1; pp. 1442 - 1452
• Main Authors: Che, Hangjun, Yang, Xuanhao
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.02.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Clustering; Clustering algorithms; Consumer electronics; Data analysis; Data models; deep matrix factorization; Factorization; Health care; Healthcare information fusion; Kernel; Matrix decomposition; Medical services; multi-kernel learning; multi-view clustering; Optimization; Task analysis
• ISSN: 0098-3063; 1558-4127
• DOI: 10.1109/TCE.2024.3440485

Multi-view clustering methods based on deep matrix factorization play a vital role in data analysis within the healthcare sector. However, existing methods predominantly conduct deep matrix factorization in the original data space, which is not conducive to addressing non-linear and complex data patterns. To address this issue, the Multi-kernel based Multi-view Deep Non-negative Matrix Factorization with Optimal Consensus Graph (OGMKMDNMF) is introduced. This approach utilizes deep non-negative matrix factorization after projecting the data matrix into a high-dimensional kernel space. Additionally, it employs optimal consensus graph to alleviate the detrimental effects arising from misassigned nearest neighbors during the construction of similarity matrix. An innovative iterative optimization algorithm is developed for OGMKMDNMF. The experimental results demonstrate the effectiveness and competitive advantage of OGMKMDNMF in addressing multi-view healthcare data clustering tasks.