Robust Discriminant Embedding Projection Fuzzy Clustering With Optimal Mean

The unsupervised nature of clustering has attracted significant interest. In particular, researchers delve into exploring the superiority of fuzzy clustering in flexibly handling computations involving uncertain data. However, outliers can present considerable challenges by distorting the measurement of similarity between samples, and biases in projection subspace learning may impede accurate partitioning. In this article, we propose a robust discriminant embedding projection fuzzy clustering with optimal mean (RPFCOM) method. First, the weighted loss function term distinguishes outliers and normal samples through boolean weight, thereby inducing row sparsity in the learning of projection subspace. The distribution of boolean weight penalizes outliers with large errors in the projection subspace. Second, we incorporate minimizing projection reconstruction information learning while suppressing redundant features, where the optimal mean dynamically corrects the projection learning bias. And the embedding of discriminative information further strengthens the capability of differentiating normal samples. Finally, the proposed method adaptively updates the boolean weight to identify outliers, which joints fuzzy membership matrix constructed from the maximum entropy graphs, enhancing the stability in distinguishing normal sample clusters. Comprehensive experimental validation on noise contaminated dataset has demonstrated the superiority of RPFCOM.