@inproceedings{d43707bf544c42e0babfdd5869f917f5,
title = "The fuzzy mega-cluster: Robustifying FCM by scaling down memberships",
abstract = "A new robust clustering scheme based on fuzzy c-means is proposed and the concept of a fuzzy mega-cluster is introduced in this paper. The fuzzy mega-cluster is conceptually similar to the noise cluster, designed to group outliers in a separate cluster. This proposed scheme, called the mega-clustering algorithm is shown to be robust against outliers. Another interesting property is its ability to distinguish between true outliers and non-outliers (vectors that are neither part of any particular cluster nor can be considered true noise). Robustness is achieved by scaling down the fuzzy memberships, as generated by FCM so that the infamous unity constraint of FCM is relaxed with the intensity of scaling differing across datum. The mega-clustering algorithm is tested on noisy data sets from literature and the results presented.",
author = "Amit Banerjee and Dav{\'e}, {Rajesh N.}",
year = "2006",
doi = "10.1007/11539506_57",
language = "English (US)",
isbn = "9783540283126",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "444--453",
booktitle = "Fuzzy Systems and Knowledge Discovery - Second International Conference, FSKD 2005, Proceedings",
address = "Germany",
note = "2nd International Conference on Fuzzy Systems and Knowledge Discovery, FSKD 2005 ; Conference date: 27-08-2005 Through 29-08-2005",
}