Feature subset-wise mixture model-based clustering via local search algorithm

Younghwan Namkoong, Yongsung Joo, Douglas D. Dankel

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In clustering, most feature selection approaches account for all the features of the data to identify a single common feature subset contributing to the discovery of the interesting clusters. However, many data can comprise multiple feature subsets, where each feature subset corresponds to the meaningful clusters differently. In this paper, we attempt to reveal a feature partition consisting of multiple non-overlapped feature blocks that each one fits a finite mixture model. To find the desired feature partition, we used a local search algorithm based on a Simulated Annealing technique. During the process of searching for the optimal feature partition, reutilization of the previous estimation results has been adopted to reduce computational cost.

Original languageEnglish
Title of host publicationAdvances in Artificial Intelligence - 23rd Canadian Conference on Artificial Intelligence, Canadian AI 2010, Proceedings
Pages135-146
Number of pages12
DOIs
StatePublished - 2010
Event23rd Canadian Conference on Artificial Intelligence, Canadian AI 2010 - Ottawa, ON, Canada
Duration: 31 May 20102 Jun 2010

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume6085 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference23rd Canadian Conference on Artificial Intelligence, Canadian AI 2010
Country/TerritoryCanada
CityOttawa, ON
Period31/05/102/06/10

Keywords

  • Clustering
  • Feature selection
  • Finite mixture model

Fingerprint

Dive into the research topics of 'Feature subset-wise mixture model-based clustering via local search algorithm'. Together they form a unique fingerprint.

Cite this