Effective fuzzy c-means clustering algorithms for data clustering problems

• Published in: Expert systems with applications Vol. 39; no. 7; pp. 6292 - 6300
• Main Authors: Kannan, S.R., Ramathilagam, S., Chung, P.C.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2012
• Subjects: Algorithms; Clustering; Data clustering; Entropy; Entropy FCM; Fuzzy; Fuzzy logic; Fuzzy set theory; Kernel distance functions; Mathematical analysis; Mathematical models; Regularization functions; Kernel distance functions; Clustering; Entropy FCM; Data clustering; Regularization functions
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2011.11.063

Clustering is a well known technique in identifying intrinsic structures and find out useful information from large amount of data. One of the most extensively used clustering techniques is the fuzzy c-means algorithm. However, computational task becomes a problem in standard objective function of fuzzy c-means due to large amount of data, measurement uncertainty in data objects. Further, the fuzzy c-means suffer to set the optimal parameters for the clustering method. Hence the goal of this paper is to produce an alternative generalization of FCM clustering techniques in order to deal with the more complicated data; called quadratic entropy based fuzzy c-means. This paper is dealing with the effective quadratic entropy fuzzy c-means using the combination of regularization function, quadratic terms, mean distance functions, and kernel distance functions. It gives a complete framework of quadratic entropy approaching for constructing effective quadratic entropy based fuzzy clustering algorithms. This paper establishes an effective way of estimating memberships and updating centers by minimizing the proposed objective functions. In order to reduce the number iterations of proposed techniques this article proposes a new algorithm to initialize the cluster centers. In order to obtain the cluster validity and choosing the number of clusters in using proposed techniques, we use silhouette method. First time, this paper segments the synthetic control chart time series directly using our proposed methods for examining the performance of methods and it shows that the proposed clustering techniques have advantages over the existing standard FCM and very recent ClusterM-k-NN in segmenting synthetic control chart time series.