Robust jointly sparse semi-supervised fuzzy C-means clustering with asymmetric deviation constraints

Area of Science:

• Data Mining
• Pattern Analysis
• Machine Learning

Background:

• Semi-supervised clustering methods leverage limited labeled data to enhance traditional fuzzy clustering.
• Existing techniques often suffer from noise sensitivity and limited performance improvements.
• Critique of current semi-supervised constraint terms in fuzzy clustering.

Purpose of the Study:

• Propose a robust jointly sparse semi-supervised fuzzy C-means clustering algorithm (RSSFCM).
• Develop an enhanced version of RSSFCM specifically for noisy data.
• Generalize existing robust and sparse fuzzy K-means clustering (RSFKM) methods.

Main Methods:

• Modified membership sparse regularization with an asymmetric deviation quadratic term.
• Algorithm convergence proven using Zangwill's theorem.
• Integration of Gaussian kernel for noisy numerical data and fuzzy local information factor for image segmentation.

Main Results:

• RSSFCM demonstrated significant performance improvements over state-of-the-art methods.
• Enhanced RSSFCM effectively handles noisy numerical data and image segmentation.
• The proposed algorithm advances semi-supervised fuzzy C-means clustering capabilities.

Conclusions:

• RSSFCM offers a robust and generalized approach to semi-supervised fuzzy clustering.
• The enhanced versions effectively address noise sensitivity in data mining applications.
• The study provides a significant advancement in the field of semi-supervised clustering.