Fuzzy nearest neighbor clustering of high-dimensional data

This paper presents a clustering method based on fuzzy nearest neighbor (FNNC algorithm for short). This method firstly finds every object´s nearest neighbor, then defines a new similarity measure which is based on the number of nearest neighbors shared by two objects, and calculates the density of every object. Next, it builds clusters with representative or core objects by eliminating noise and associating non-noise objects. FNNC algorithm relies on weight shared nearest neighbor graph, when it builds clusters, it only uses some useful links between objects of the graph. The experiments show that FNNC algorithm can efficiently find clusters in high-dimensional data space.