MDL-Based Selection of the Number of Components in Mixture Models for Pattern Classification

  title={MDL-Based Selection of the Number of Components in Mixture Models for Pattern Classification},
  author={Hiroshi Tenmoto and Mineichi Kudo and Masaru Shimbo},
A new method is proposed for selection of the optimal number of components of a mixture model for pattern classiication. We approximate a class-conditional density by a mixture of Gaussian components. We estimate the parameters of the mixture components by the EM (Expectation Maximization) algorithm and select the optimal number of components on the basis of the MDL (Minimum Description Length) principle. We evaluate the goodness of an estimated model in a trade-oo between the number of the… CONTINUE READING
Highly Cited
This paper has 25 citations. REVIEW CITATIONS
16 Citations
9 References
Similar Papers


Publications citing this paper.
Showing 1-10 of 16 extracted citations


Publications referenced by this paper.
Showing 1-9 of 9 references

Robust Clustering Based on a Maximum Likelihood Method for Esti mation of the Suitable Number of Clusters

  • N. Ichimura
  • Trans. Inst. Electron. Info. Comm. Eng. D-II
  • 1995
Highly Influential
5 Excerpts

Automated Design of Multiple-Class Piecewise Linear Clas si ers

  • Y. Park, J. Sklansky
  • J. Classi cation 6
  • 1989
2 Excerpts

Mixture Models

  • G. J. McLachlan, K. E. Basford
  • Marcel Dekker, Inc., New York
  • 1988
1 Excerpt

Generation of VCV/CVC Balanced Word Sets for Speech Database

  • S Hayamizu
  • Bullet. Electro. Lab
  • 1985
2 Excerpts

A Universal Prior for Integers and Estimation by MinimumDescription Length

  • J. Rissanen
  • Ann. Stat. 11
  • 1983
2 Excerpts

Cluster Validity with Fuzzy Sets

  • J. C. Bezdek
  • J. Cybern
  • 1974
1 Excerpt

Similar Papers

Loading similar papers…