Ahmed, Denez Hydir (2026) Semi-Supervised Feature and View Weighted Multi-View Fuzzy C-Means Clustering under Weak Supervision. CENTRAL ASIAN JOURNAL OF MATHEMATICAL THEORY AND COMPUTER SCIENCES, 7 (4). ISSN 2660-5309

Text CAJMTCS_Semi-Supervised+Feature.pdf Download (566kB)

Abstract

Fuzzy C-Means assumes all features are equally informative, all views are equally informative, and it does not use any available labeled data. In real multi-view data, some views may contain the cluster structure quite explicitly, while others may not provide any information other than noise, and a small subset of the data may be labeled, albeit imperfectly, but may also suggest the right answer if the algorithm takes this into account. This proposal introduces SSFVW-MVFCM, a single objective function that weights features within each view, weights the views and includes a term corresponding to weak partial supervision which is updated in conjunction with the other terms through block coordinate descent with closed-form updates at each step. A confidence mechanism, provided by a brief unsupervised warm-start, mitigates the over-trust of labelled samples whose label is different from their feature space position. Numerous experiments are performed, with the model tested on three real multi-view datasets using UCI-sourced data: handwritten digits, breast-tumor diagnostics, and wine chemistry, a comparison with five baseline methods, an investigation of the number of views, an investigation of view-quality imbalance, an investigation of feature-weighting and view-weighting ablations, an investigation of labeled-data ratio, and an investigation of robustness to label noise. Evaluation is based on ACC, NMI, ARI, Purity, and pairwise F-measure and is given as a mean along with a standard deviation over multiple runs and is then statistically tested using the paired Wilcoxon test. The preliminary results indicate that the proposed model outperforms all the unsupervised baseline methods by a large margin on all three datasets, with the most significant contributions coming from the partial-supervision term, and not weighting. But its superiority over a single-view semi-supervised version was not ubiquitous, the proposal reports; multiple views helped on one dataset and hurt on two others.

Item Type:	Article
Uncontrolled Keywords:	Multi-view clustering, Fuzzy C-Means, feature weighting, view weighting, weak supervision
Subjects:	H Social Sciences
Depositing User:	admin eprints
Date Deposited:	10 Oct 2026 04:45
Last Modified:	10 Oct 2026 04:45
URI:	http://eprints.umsida.ac.id/id/eprint/17160

Actions (login required)

View Item