Earticle

초록

영어

The kmeans algorithm is an unsupervised classification algorithm. It has some drawbacks, the number of classes has to be known a priori, the initialization phase and the local optimums. We present in this paper an improvement based on evolutionary strategies and on the Xie and Beni criterion in order to get around these three difficulties. We design a new evolutionist kmeans algorithm. We suggest a new mutation operator that allows the algorithm to avoid local solutions and to converge to the global solution in a small computation time. We have optimized the Xie and Beni criterion by evolutionary strategies for the optimal choice of the number of classes. The proposed method is validated on several simulation examples. The experimental results obtained show the rapid convergence and the good performances of this new approach.

목차

Abstract
 1. Introduction
 2. Evolutionary strategies
 3. kmeans classification
  3.1. Descriptive elements
  3.2. kmeans algorithm
 4. Evolutionary kmeans classification
  4.1. Proposed coding
  4.2. The proposed fitness function
  4.3. The proposed mutation operator
  4.4. The proposed EKM algorithm
 5. Determination of the optimal number of classes
  5.1. Xie and Beni criterion
  5.2 Optimization of Xie and Beni criterion by evolutionary strategies
 6. Experimental results and evaluations
  6.1. Introduction
  6.2. Test 1
  6.3. Test 2
  6.4. Test 3
  6.5. Test 4
  6.6. Estimation of the optimal number of classes
 7. Conclusion
 8. References

키워드

저자

• M. Merzougui [ LABO MATSI, EST, University of Mohammed ]
• A. EL Allaoui [ LABO MATSI, EST, University of Mohammed ]
• M. Nasri, M. EL Hitmy [ LABO MATSI, EST, University of Mohammed ]
• H. Ouariachi [ LABO MATSI, EST, University of Mohammed ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  International Journal of Advanced Science and Technology

• 간기

  월간

• pISSN

  2005-4238

• 수록기간

  2008~2016

• 십진분류

  KDC 505 DDC 605