Advertisement

Local Bayesian Based Rejection Method for HSC Ensemble

  • Qing He
  • Wenjuan Luo
  • Fuzhen Zhuang
  • Zhongzhi Shi
Part of the Lecture Notes in Computer Science book series (LNCS, volume 6063)

Abstract

Based on Jordan Curve Theorem, a universal classification method, called Hyper Surface Classifier (HSC) was proposed in 2002. Experiments showed the efficiency and effectiveness of this algorithm. Afterwards, an ensemble manner for HSC(HSC Ensemble), which generates sub classifiers with every 3 dimensions of data, has been proposed to deal with high dimensional datasets. However, as a kind of covering algorithm, HSC Ensemble also suffers from rejection which is a common problem in covering algorithms. In this paper, we propose a local bayesian based rejection method(LBBR) to deal with the rejection problem in HSC Ensemble. Experimental results show that this method can significantly reduce the rejection rate of HSC Ensemble as well as enlarge the coverage of HSC. As a result, even for datasets of high rejection rate more than 80%, this method can still achieve good performance.

Keywords

HyperSurface Classification (HSC) HSC Ensemble Rejection 

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    Furnkranz, J.: ROC n Rule Learning Towards a Better Understanding of Covering Algorithms. Machine Learning 58, 39–77 (2005)CrossRefGoogle Scholar
  2. 2.
    Zhang, L., Zhang, B.: A Geometrical Representation of McCullochCPitts Neural Model and Its Applications. IEEE Transactions on Neural Networks 10(4) (1999)Google Scholar
  3. 3.
    Wu, T., Zhang, L., Yan-Ping, Z.: Kernel Covering Algorithm for Machine Learning. Chinese Journal of Computers 28(8) (2005)Google Scholar
  4. 4.
    He, Q., Shi, Z.-Z., Ren, L.-A., Lee, E.S.: A Novel Classification Method Based on Hyper Surface. International Journal of Mathematical and Computer Modeling, 395–407 (2003)Google Scholar
  5. 5.
    Pan, S.J., Yang, Q.: A Survey on Transfer Learning. IEEE Transactions on Knowledge and Data Engineering (October 12, 2009)Google Scholar
  6. 6.
    Landgrebe, T.C.W., Tax, D.M.J., Paclk, P., Duin, R.P.W.: The interaction between classification and reject performance for distance-based reject-option classifiers. Pattern Recognition Letters 27(8), 908–917 (2006)CrossRefGoogle Scholar
  7. 7.
    Dubuisson, B., Masson, M.: A statistical decision rule with incomplete knowledge about classes. Pattern Recognition 26(1), 155–165 (1993)CrossRefGoogle Scholar
  8. 8.
    Landgrebe, T., Tax, D., Paclk, P., Duin, R., Andrew, C.: A combining strategy for ill-defined problems. In: Fifteenth Ann. Sympos. of the Pattern Recognition Association of South Africa, pp. 57–62 (2004)Google Scholar
  9. 9.
    Zhang, L., Wu, T., Zhou, Y., Zhang, Y.P.: Probabilistic Model for Covering Algorithm. Journal of Software 18(11), 2691–2699 (2007)zbMATHCrossRefGoogle Scholar
  10. 10.
    He, Q., Zhao, X., Shi, Z.: Classification based on dimension transposition for high dimension dataInternational Journal Soft Computing-A Fusion of Foundations. Methodologies and Applications, 329–334 (2006)Google Scholar
  11. 11.
    Zhao, X.R., He, Q., Shi, Z.Z.: HyperSurface Classifiers Ensemble for High Dimensional Data sets. In: Wang, J., Yi, Z., Żurada, J.M., Lu, B.-L., Yin, H. (eds.) ISNN 2006. LNCS, vol. 3971, pp. 1299–1304. Springer, Heidelberg (2006)CrossRefGoogle Scholar
  12. 12.
    He, Q., Zhao, X.-R., Shi, Z.-Z.: Minimal consistent subset for hyper surface classification method. International Journal of Pattern Recognition and Artificial Iintelligence 22(1) (2008)Google Scholar
  13. 13.

Copyright information

© Springer-Verlag Berlin Heidelberg 2010

Authors and Affiliations

  • Qing He
    • 1
  • Wenjuan Luo
    • 1
    • 2
  • Fuzhen Zhuang
    • 1
    • 2
  • Zhongzhi Shi
    • 1
  1. 1.The Key Laboratory of Intelligent Information Processing, Institute of Computing TechnologyChinese Academy of SciencesBeijingChina
  2. 2.Graduate University of Chinese Academy of SciencesBeijingChina

Personalised recommendations