An enhanced sequential fuzzy clustering algorithm

Abstract

A sequential fuzzy clustering algorithm is proposed based on a modification to the objective function used in the fuzzy competitive learning algorithm. The new learning algorithm can be used to enhance the excitation on the non-winning centroids and to reduce the excitation on the winning centroid when the fuzziness parameter is close to unit. The excitation on the winning centroid can be further reduced when the input pattern is far away from the winning centroid. An excitationinhibition mechanism can also be introduced into the learning such that the non-winning centroids move towards the input pattern while the winning centroid moves away from the input pattern when the winning centroid is far away from the input pattern. The new algorithm overcomes the problem of underutilization of centroids found in the k -means or related clustering algorithms and in the fuzzy competitive learning algorithm when the fuzziness parameter is close to unity. The performance of the new algorithm is demonstrated on the IRIS data set.