Ann. Phys. Fr.
Volume 20, Number 1, 1995
|1 - 45
|01 June 2004
Multivariate Analysis in Nuclear Physics*
Institut des Sciences Nucléeaires, IN2P3-CNRS et Universitée Joseph Fourier, 53 Avenue des Martyrs, 38026 Grenoble Cedex, France
Nuclear physics deals more and more with experiments involving a large number of parameters. The analysis of such experiments requires well adapted statistical techniques. The multivariate analysis techniques consist in the representation of the experimental events as points in the multidimensional space of the physical variables. One aim will be to treat experimental information as a whole. This formalism permits the simultaneous studies of the structures of the event cloud and of the correlations between the variables. Principal Component Analysis is concerned with the determination of the so-called principal variables, linear combinations of the primary physical variables, which represent the maximum information. Correspondence Analysis visualises, on a 2D diagram, the correlations between the modalities of qualitative variables. The goal of the Discriminant Analysis is to discriminate different types of events, that is to affect them to a familly. The last part of the work is devoted to a global protocol, involving the PCA, for the comparison of experimental data with data generated by a simulation code.
© EDP Sciences, 1995