Advertisement

Stratification Before Discriminant Analysis: A Must?

  • Jean-Paul Rasson
  • Jean-Yves Pirçon
  • François Roland
Part of the Studies in Classification, Data Analysis, and Knowledge Organization book series (STUDIES CLASS)

Abstract

It could be said as a tautology that, if we want to make a discriminant analysis between two or more populations and if we are able to divide these populations and training sets into some homogeneous subsets, it will be more efficient to make it on each of these subsets and then to combine the results. This can be done using one or two variables highly correlated with the one we want to predict. Our point of view will be a bit different: we will use a classification tree on all the available variables. We will first recall the first attempt (presented at IFCS2002 in Krakow). This one allowed us to obtain on an example of prediction of failure of the enterprises a gain of 5% of well classified data, using, after and before stratification, the classical Fisher’s linear discriminant rule or the logistic regression. We intend to present a new method, still a classification tree, but with a multivariate criterion and in an agglomerative way. We compare both methods. In the same conditions and with the same data set, the gain is as high as 20%! Results will obviously also be presented when the methods are applied to test sets. Finally, we will conclude.

Keywords

Discriminant Analysis Poisson Process Classification Tree Normal Kernel Homogeneous Poisson Process 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. BOCK, H.H. (1989): Probabilistic Aspects in Cluster Analysis. Conceptual and Numerical Analysis of Data, 12–44.Google Scholar
  2. BREIMAN, L., FRIEDMAN, J.H., OLSHEN, R.A., and STONE, C.J. (1984): Clas-sification and Regression Trees. Belmont, Wadsworth.Google Scholar
  3. CHAVENT, M. (1997): Analyse des données symboliques. Une méthode divisive de classification. PhD thesis. Universit de Paris IX, Dauphine.Google Scholar
  4. DAUDIN, J-J., MASSON, J-P., TOMASSONE, R., and DANZART, M. (1988): Discrimination et classement. Masson, Paris.Google Scholar
  5. SILVERMAN, B.W. (1986): Density Estimation for Statistics and Data Analysis. Chapman and Hall, London.Google Scholar
  6. WILLIAMS, W.T. and LAMBERT, J.M. (1959): Multivariate Method in Plant Ecology. Journal of Ecology, 47, 83–101.Google Scholar

Copyright information

© Springer-Verlag Berlin · Heidelberg 2005

Authors and Affiliations

  • Jean-Paul Rasson
    • 1
  • Jean-Yves Pirçon
    • 1
  • François Roland
    • 1
  1. 1.Department of MathematicsUniversity of NamurNamurBelgium

Personalised recommendations