Outlier Resistant Fuzzy Clustering via Row Sparse Discriminative Embedding Projection

Abstract

Fuzzy clustering and its derivatives have been widely applied for handling overlapping clusters throughprobabilistic membership assignment, yet their performance degrades under cumulative outlier interference. To cope with this limitation, we propose the Outlier Resistant Fuzzy Clustering via Row Sparse Discriminative Embedding Projection (RFCDE), which introduces an adaptive sample contribution vector to resist the outliers, a row-sparse membership refinement strategy to enhance normal sample attention, and a projection-guided prototype learning module to mitigate representation bias. Furthermore, a discriminative embedding objective is designed to effectively mitigate extraneous feature effects. These modules form a unified iterative architecture that improves clustering reliability in a low-dimensional framework. Comparative experiments on real-world datasets validate its broad applicability.