Advertisement

TrueSkill-Based Pairwise Coupling for Multi-class Classification

  • Jong-Seok Lee
Part of the Lecture Notes in Computer Science book series (LNCS, volume 7553)

Abstract

A multi-class classification problem can be solved efficiently via decomposition of the problem into multiple binary classification problems. As a way of such decomposition, we propose a novel pairwise coupling method based on the TrueSkill ranking system. Instead of aggregating all pairwise binary classification results for the final decision, the proposed method keeps track of the ranks of the classes during the successive binary classification procedure. Especially, selection of a binary classifier at a certain step is done in such a way that the multi-class classification decision using the binary classification results up to the step converges to the final one as quickly as possible. Thus, the number of binary classifications can be reduced, which in turn reduces the computational complexity of the whole classification system. Experimental results show that the complexity is reduced significantly for no or minor loss of classification performance.

Keywords

TrueSkill multi-class classification classifier fusion match-making pairwise coupling on-line ranking 

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    Allwein, E.L., Schapire, R.E., Singer, Y.: Reducing multiclass to binary: a unifying approach for margin classifiers. J. Mach. Learn. Res. 1, 113–141 (2001)MATHMathSciNetGoogle Scholar
  2. 2.
    Chang, C.C., Lin, C.J.: LIBSVM: a library for support vector machines. ACM Trans. Intell. Syst. Techn. 2(3), 27:1–27:27 (2011), http://www.csie.ntu.edu.tw/~cjlin/libsvm Google Scholar
  3. 3.
    Crammer, K., Singer, Y.: On the algorithmic implementation of multiclass kernel-based vector machines. J. Mach. Learn. Res. 2, 265–292 (2001)Google Scholar
  4. 4.
    Dietterich, T.G., Bakiri, G.: Solving multiclass learning problems via error-correcting output codes. J. Artif. Intell. Res. 2, 263–286 (1995)MATHGoogle Scholar
  5. 5.
    Elo, A.E.: The Rating of Chessplayers, Past and Present. Arco Publishing, New York (1986)Google Scholar
  6. 6.
    Frank, A., Asuncion, A.: UCI machine learning repository. School of Information and Computer Science, University of California, Irvine (2010)Google Scholar
  7. 7.
    García-Pedrajas, N., Ortiz-Boyer, D.: An empirical study of binary classifier fusion methods for multiclass classification. Inform. Fusion 12, 111–130 (2011)CrossRefGoogle Scholar
  8. 8.
    Glickmann, M.E.: Parameter estimation in large dynamic paired comparison experiments. Appl. Statist. 48, 377–394 (1999)Google Scholar
  9. 9.
    Hastie, T., Tibshirani, R.: Classification by pairwise coupling. Ann. Stat. 26(2), 451–471 (1998)CrossRefMATHMathSciNetGoogle Scholar
  10. 10.
    Herbrich, R., Minka, T., Graepel, T.: Trueskill: a Bayesian skill rating system. In: Adv. Neural Info. Process. Syst., vol. 19, pp. 569–576. MIT Press, Cambridge (2007)Google Scholar
  11. 11.
    Hsu, C.W., Lin, C.J.: A comparison of methods for multiclass support vector machines. IEEE Trans. Neural Netw. 13(2), 415–425 (2002)CrossRefGoogle Scholar
  12. 12.
    Kreßel, U.: Pairwise classification and support vector machines. In: Schölkopf, B., Burges, C.J.C., Smola, A.J. (eds.) Advances in Kernel Methods-Support Vector Learning, pp. 255–268. MIT Press, Cambridge (1999)Google Scholar
  13. 13.
    Lorena, A.C., de Carvalho, A.C.P.L.F., Gama, J.M.P.: A review on the combination of binary classifiers in multiclass problems. Artf. Intell. Rev. 30, 19–37 (2008)CrossRefGoogle Scholar
  14. 14.
    Rifkin, R., Klautau, A.: In defense of one-vs-all classification. J. Mach. Learn. Res. 5, 101–141 (2004)MATHMathSciNetGoogle Scholar
  15. 15.
    Weng, R.C., Lin, C.J.: A Bayesian approximation method for online ranking. J. Mach. Learn. Res. 12, 267–300 (2011)MATHMathSciNetGoogle Scholar
  16. 16.
    Wu, T.F., Lin, C.J., Weng, R.C.: Probability estimates for multi-class classification by pairwise coupling. J. Mach. Learn. Res. 5, 975–1005 (2004)MATHMathSciNetGoogle Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2012

Authors and Affiliations

  • Jong-Seok Lee
    • 1
  1. 1.School of Integrated TechnologyYonsei UniversityKorea

Personalised recommendations