A novel method for identifying optimal number of clusters with marginal differential entropy

  • Bo Shu
  • , Wei Chen*
  • , Zhendong Niu
  • , Changmin Zhang
  • , Xiaotian Jiang
  • *Corresponding author for this work

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

Abstract

Clustering evaluation plays an important role in clustering algorithms. Most of recent approaches about clustering that evaluate and identify the optimal number of clusters need to calculate the distances between data points pair-wisely or evaluate the entropy in the entire dimension space and have high computational complexity. In this paper, we propose an entropy-based clustering evaluation method for identifying the optimal number of clusters which first projects the clusters centroids to each of its individual dimensions, then accumulates the marginal differential entropy in each dimension. With the sum of marginal entropies we can analyze the performance and identify the optimal number of clusters. This method can dramatically reduce the computational complexity without losing accuracy. Experiment results show that the proposed method has high stability under various situations and can apply to massive high-dimensional data points.

Original languageEnglish
Title of host publicationWeb-Age Information Management - WAIM 2013, International Workshops
Subtitle of host publicationHardBD, MDSP, BigEM, TMSN, LQPM, BDMS, Proceedings
Pages371-382
Number of pages12
DOIs
Publication statusPublished - 2013
Event14th International Conference on Web-Age Information Management, WAIM 2013 - Beidaihe, China
Duration: 14 Jun 201316 Jun 2013

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume7901 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference14th International Conference on Web-Age Information Management, WAIM 2013
Country/TerritoryChina
CityBeidaihe
Period14/06/1316/06/13

Keywords

  • Clustering Evaluation
  • Differential Entropy
  • Information Theory

Fingerprint

Dive into the research topics of 'A novel method for identifying optimal number of clusters with marginal differential entropy'. Together they form a unique fingerprint.

Cite this