An Integrity-based Fuzzy C-means Clustering Algorithm Conforming to Human Perception

Abstract

Traditional fuzzy c-means (FCM) that uses a least squared errors function as its objective function tends to partition a dataset to clusters of equal population for obtaining a mathematically optimal solution; however, such results are not satisfactory for cases of unequal clusters from human perception. csiFCM introduces a ratio of cluster sizes as the condition value to balance the influence of different clusters, but its performance is predominantly depending on the initial center positions. In order to produce clustering results conforming to human perception even for unequal size cases, we propose an integrity-based FCM method, which determines the contribution of each data by considering the data relationship of both inter- and intra- clusters, to improve the insufficiency of conditional FCM on solving quantity variation problem. Experiment results show that our proposed integrity-based FCM clustering method not only can suppress the influence of dataset in larger clusters effectively, but also can update the wrong initial centers to the positions conforming to human perception.