Locality sensitive C-means clustering algorithms

The concept of preserving locality information in dimensionality reduction and semi-supervised classification have been very popular recently. In this paper, we attempt to use locality sensitive weight for clustering, where the neighborhood structure information between objects are transformed into weights of objects. We develop two novel locality sensitive C-means algorithms, i.e. Locality-weighted Hard C-Means (LHCM) and Locality-weighted Fuzzy C-Means (LFCM), following the standard C-Means and fuzzy C-means, respectively. In addition, two semi-supervised extensions of LFCM are proposed to better use some given partial supervision information in data objects. Experimental results on both artificial and real datasets validate the effectiveness of the proposed algorithms.