Robust and Sparse Fuzzy K-Means Clustering

The partition-based clustering algorithms, like KMeans and fuzzy K-Means, are most widely and successfully used in data mining in the past decades. In this paper, we present a robust and sparse fuzzy K-Means clustering algorithm, an extension to the standard fuzzy K-Means algorithm by incorporating a robust function, rather than the square data fitting term, to handle outliers. More importantly, combined with the concept of sparseness, the new algorithm further introduces a penalty term to make the object-clusters membership of each sample have suitable sparseness. Experimental results on benchmark datasets demonstrate that the proposed algorithm not only can ensure the robustness of such soft clustering algorithm in real world applications, but also can avoid the performance degradation by considering the membership sparsity.