Adaptive weighted fuzzy clustering based on intra-cluster data divergence

Fuzzy C-means clustering (FCM) approach is an effective method for clustering and has been successfully applied in numbers of real-world problems. In this paper, we propose an improving adaptive weighted FCM based on data divergence, with the merits of three aspects: 1) to avoid randomization of cluster centers, we propose a new cluster centers initialization method; 2) we present an adaptive parameter reflecting the changes of intra-cluster data divergence in the process of cluster formation from iteration to iteration for correcting the unreasonable factors resulting from the changes timely; 3) we propose a new data weighting method. By integrating the adaptive parameter and feature weighting method, we propose a novel adaptive objective function, by which the updating iterative formulas of the membership degrees, the feature weights and the cluster centers are obtained. Experimental results have shown that the novel clustering approach put forward can improve the clustering performance effectively.