Robust Sparse Twin Support Vector Clustering via ϵ-Insensitive Margins

Clustering is a fundamental machine learning technique for analyzing large volumes of unlabeled data. Constructing accurate cluster boundaries that remain resilient to feature noise without sacrificing model sparsity is a primary challenge in complex environments. Traditional twin support vector clustering algorithms rely on the hinge loss, which is highly unstable under noise corruption and lacks re-sampling stability. By incorporating an ϵ-insensitive pinball loss function, a controlled penalty mechanism is created that ignores errors within a specific margin while bounding the penalty for outliers. The Sparse Pinball loss Twin Support Vector Clustering framework is developed to simultaneously balance noise robustness and structural sparsity. Extensive evaluations demonstrate that this approach achieves superior clustering accuracy and significantly improved testing efficiency on both synthetic benchmarks and real-world biomedical datasets.