Clustering by Unified Principal Component Analysis and Fuzzy C-Means with Sparsity Constraint

For clustering high-dimensional data, most of the state-of-the-art algorithms often extract principal component beforehand, and then conduct a concrete clustering method. However, the two-stage strategy may deviate from assignments by directly optimizing the unified objective function. Different from the traditional methods, we propose a novel method referred to as clustering by unified principal component analysis and fuzzy c-means (UPF) for clustering high-dimensional data. Our model can explore underlying clustering structure in low-dimensional space and finish clustering simultaneously. In particular, we impose a L0-norm constraint on the membership matrix to make the matrix more sparse. To solve the model, we propose an effective iterative optimization algorithm. Extensive experiments on several benchmark data sets in comparison with two-stage algorithms are conducted to validate effectiveness of the proposed method. The experiments results demonstrate that the performance of our proposed method is superiority.