Shadowed C-means clustering based on approximated feature space

The random Fourier Features method has been found very effective in approximating the kernel functions. Our former studies show that through a mixing mechanism of the feature space formed by random Fourier features and certain linear algorithms, the fuzzy clustering results in the approximated feature space are comparable to or even exceed the classical kernel-based algorithms. To increase the robustness of clustering results over outliers, this paper proposes to employ the shadowed C-Means algorithm in the approximated kernel feature space generated by the random Fourier Features method. The experiments compared with traditional fuzzy C-Means algorithm demonstrate the strengths and efficiency of the proposed approach.