Entropy regularization for sparse robust fuzzy clustering via boolean weighting

Fuzzy clustering is a fundamental tool for unsupervised pattern recognition, yet its performance inevitably degrades when dealing with data contaminated by noise and outliers. Conventional robust extensions typically rely on continuous soft weighting schemes. However, under severe contamination, these methods fail to fully exclude extreme anomalies, allowing residual biased gradients to accumulate and cause unrecoverable centroid drift. To address this limitation, we propose Entropy Regularization for Sparse Robust Fuzzy Clustering via boolean weighting (ERSRFC). Instead of assigning continuous contributions to all samples, the proposed method determines their influence according to the consistency with evolving cluster structure, so that unreliable observations are progressively suppressed during optimization. This mechanism is integrated with a ℓ_2,p-norm distance measure and an entropy based regularization strategy, which together regulate sample participation, residual influence, and membership stability in a unified manner. Extensive experiments demonstrate that ERSRFC achieves stable and competitive clustering performance compared to existing advanced methods.