Weighted possibilistic c-means clustering algorithms

This paper proposes the weighted possibilistic c-means algorithm. The weights indicate the possibility of a given feature vector belongs to any cluster. By assigning low weight values to outliers, the effects of noisy data on the clustering process is reduced. It is shown that the possibilistic c-means algorithm is a special case of the weighted possibilistic c-means algorithm if each feature vector weight is assigned to one. Several methods for determining the weight values are presented. The performance of the algorithms is tested using data generated by a Gaussian random number generator with outliers and an artificial data set containing outliers