Linear and non-linear discriminant functions for the classification of high energy physics data
by
Mostafa Mjahed
Maths and Systems Department, Ecole Royale de l'Air, 40000 Marrakech, Morocco

The High Energy Physics data are noisy and the constituent classes strongly overlap. We propose, in this paper, to use multivariate analysis methods as linear and non-linear discriminant analysis in order to improve the separation between several kinds of events. Global event shape and connected variables are considered. The same sets of training and test data, containing four classes, are employed for the two methods. A test of performance using efficiencies and purities of classifications is used.

The main purpose of a linear discriminant function analysis is to predict class membership based on a linear combination of the considered variables. This can be represented as a use of multi-dimensional hyper-planes for the separation. A non-linear discriminant classifier is achieved by generalizing the linear discriminant functions to non-linear hyper-surfaces. The extension of linear and non linear discriminant function analysis to situations with four classes is straightforward. Due to its non linearity, the non-linear discriminant classifier cope best with the complexity of the considered classes shapes. This method is correct in more than 89 percent of the generated events.

References

[1] G. Saporta, Probabilités, Analyse des Données et Statistique, Editions Technip, Paris, 1990.

[2] M. S. Srivastava, E. M. Carter, Applied multivariate statistics, North Holland, Amsterdam, 1983.

[3] A. J. Dobson, An introduction to generalized linear models, Chapman Hall, New York, 1990.

Date received: July 16, 2001

Copyright © 2001 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cahv-72.