Relative Density Weights Based Fuzzy C-Means Clustering Algorithms

Fuzzy C-means (FCM) clustering algorithm tries to get the memberships of each sample to each Cluster by optimizing an objective function, and then assign each of the samples to an appropriate class. The Fuzzy C-means algorithm doesn't fit for clusters with different sizes and different densities, and it is sensitive to noise and anomaly. We present two improved fuzzy c-means algorithms, Clusters-Independent Relative Density Weights based Fuzzy C-means (CIRDWFCM) and Clusters-Dependent Relative Density Weights based Fuzzy C-means (CDRDWFCM), according to the various roles of different samples in clustering. Several experiments of them are done on four datasets from UCI and UCR. Experimental results shows that this two presented algorithms can increase the similarity or decrease the iterations to some extent, and get better clustering results and improve the clustering quality.