Clustering refinement

Félix Iglesias*, Tanja Zseby, Arthur Zimek

*Corresponding author for this work

Research output: Contribution to journalJournal articleResearchpeer-review

Abstract

Advanced validation of cluster analysis is expected to increase confidence and allow reliable implementations. In this work, we describe and test CluReAL, an algorithm for refining clustering irrespective of the method used in the first place. Moreover, we present ideograms that enable summarizing and properly interpreting problem spaces that have been clustered. The presented techniques are built on absolute cluster validity indices. Experiments cover a wide variety of scenarios and six of the most popular clustering techniques. Results show the potential of CluReAL for enhancing clustering and the suitability of ideograms to understand the context of the data through the lens of the cluster analysis. Refinement and interpretability are both crucial to reduce failure and increase performance control and operational awareness in unsupervised analysis.

Original languageEnglish
JournalInternational Journal of Data Science and Analytics
Volume12
Issue number4
Pages (from-to)333-353
ISSN2364-415X
DOIs
Publication statusPublished - Oct 2021

Bibliographical note

Publisher Copyright:
© 2021, The Author(s).

Keywords

  • Cluster refinement
  • Cluster validity
  • Machine learning interpretability

Fingerprint

Dive into the research topics of 'Clustering refinement'. Together they form a unique fingerprint.

Cite this