Mutual information based weighted fuzzy clustering
Yogalakshmi Jayabal, Chandrashekar Ramanathan · 2014
Fuzzy C-Means (FCM) is one of the most widely used fuzzy clustering algorithm and finds its application in a variety of domains. There exists huge number of works on variations of FCM such as Weighted FCM, FCM based on different distance metrics, FCM with different clustering criterion etc. This paper studies weighted FCM. Most of weighted FCM are based on Lagrange methods of weighting either memberships of objects or attributes. These methods estimate weights based on similarity between different data objects. Mutual information, an information theoretic based measure, estimates, how much one random variable tells us about another. It can be thought of as the reduction in uncertainty about one random variable given knowledge of another. There are many variations of fuzzy c-means clustering approaches available using mutual information or variations of mutual information as a distance metric for finding the coherent clusters. In this paper, we explore this mutual information for determining feature weights. We present weighted FCM, where weighting is based on mutual information between two variables. From the experiments, it is seen that the proposed method performs better in terms of accuracy and finds a reasonable structure in data, which is evident from the calculated average silhouette width. Additionally, the evaluated kappa co-efficient of the clusters also show the improvement achieved by the proposed method.