Clustering based on possibilistic entropy

Herein we present a new clustering technique within the framework of possibilistic theory First, the possibilistic entropy is defined with brief discussion. Then the Possibilistic Entropy Clustering (PEC) algorithm is developed, which is of clear physical meaning and well-defined mathematical features and takes into account both global effect and local effect of entropy based clustering. Besides, it can automatically control the resolution parameter during the clustering proceeds and overcome the sensitivity to noise and outliers. Finally, illustrative examples show that this novel algorithm provides efficient and robust estimation of the prototype parameters even when the clusters vary significantly in size and shape, and the data set is contaminated by heavy noise.