A heuristic adjustment to the calculation of the dissimilarity in the FCM algorithm

In this paper, one of the most widely used fuzzy clustering model, fuzzy c-means (FCM) is discussed. The FCM algorithm is based on the sum of intracluster distances criterion. This criterion is effective only when the data set contains clusters that are well-separated or have similar shape and volume. In order to minimize the objective function of the FCM algorithm, the small clusters grab some points belonging to the largest clusters. This article presents a simple and intuitive idea to approach this problem. It consists of some heuristic adjustments to the calculation of the Euclidean distances employed in FCM algorithm. The heuristics change the distances from the points to the prototypes, based on the size and the orientation of the clusters. Benefits of the methodology are illustrated in the results of the simulations carried out using some artificial data sets