The fuzzy mega-cluster: Robustifying FCM by scaling down memberships

Amit Banerjee, Rajesh Dave

Research output: Chapter in Book/Report/Conference proceedingConference contribution

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.

Original languageEnglish (US)
Title of host publicationFuzzy Systems and Knowledge Discovery - Second International Conference, FSKD 2005, Proceedings
PublisherSpringer Verlag
Pages444-453
Number of pages10
Volume3613 LNAI
ISBN (Print)9783540283126
StatePublished - Jan 1 2006
Event2nd International Conference on Fuzzy Systems and Knowledge Discovery, FSKD 2005 - Changsa, China
Duration: Aug 27 2005Aug 29 2005

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume3613 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other2nd International Conference on Fuzzy Systems and Knowledge Discovery, FSKD 2005
CountryChina
CityChangsa
Period8/27/058/29/05

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'The fuzzy mega-cluster: Robustifying FCM by scaling down memberships'. Together they form a unique fingerprint.

Cite this