GFCM: A gossip-based fuzzy C-means algorithm

In real-world networks, large amounts of data are distributed among nodes. Clustering is a useful method for analyzing large data sets and the Fuzzy C-Means (FCM) algorithm is one of the most popular methods for fuzzy clustering. In this paper, we propose GFCM, a gossip-based fuzzy C-means algorithm, for collective discovery of a clustering model from data residing at individual data sites. The proposed algorithm consists of local and collaborative phases. Nodes continuously collaborate with each other to maintain a summarized view of the data set, and to find the correct fuzzy clustering model for their local data. Gossiping is used as a robust method for communication between nodes. The experimental results show that GFCM can detect fuzzy clusters efficiently, with bounded communication costs. Its performance is also compared the recently proposed CFCM algorithm.