.png){kind=icon}
Return
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
Abstract
En 中文
Clustering in noisy and complex environments remains challenging. Conventional distance measures often fail because they are sensitive to noise, feature scales, and nonlinear structures. This paper proposes a robust clustering method based on multi-granularity difference relations and shadow set theory, referred to as CMGS. The approach avoids traditional metric distances. It characterizes object similarity through attribute-wise differences, creating a finer-grained, noise-tolerant representation of data relationships. Shadow set theory models uncertainty in these differences. Adaptive thresholds learned via gradient descent suppress noise and mitigate scale effects across attributes. Integrating single-attribute and global difference granularities creates a novel similarity and distance measure for clustering. Extensive experiments on real-world and synthetic datasets validate the method. CMGS consistently outperforms classical and recent clustering methods in accuracy, stability, and robustness. It proves particularly effective in high-dimensional, imbalanced, and noisy scenarios. The framework provides a principled and effective solution for robust clustering under uncertainty.
Keywords:
Clustering
Shadow Set Theory
Multi-Granularity
Difference Relations
Noisy Data
Journal
IF:
2.7
Papers:
7.6K
Citations:
1.5W
Organization
Cited Papers
No cited papers available