A novel fuzzy entropy clustering algorithm

This paper presents a novel unsupervised clustering algorithm for data classification using a fuzzy entropy approach. It is well known that the performances of conventional objective optimization algorithms, like k-means and fuzzy c-means (FCM), etc., heavily depend on priori information, such as the number of clusters. Here, the authors propose a new type of clustering algorithm which is developed by thresholding the object's distance matrix and its neighborhood association. The proposed algorithm has superiority over conventional algorithms when the number and the shape of clusters are hard to obtain and the solution sticks to a local optimal solution. In theory, the algorithm can be applied to clusters of arbitrary shape. The algorithm has been applied to the data which are either spherical or linear in shape.<>