Classification of the HIPPARCOS Input Catalogue Using the Kohonen Network

Abstract

The unsupervised classification with respect to stellar populations of 106 000 of the 118 000 Hipparcos input catalogue stars is presented, in terms of angular position and velocity, apparent magnitude and spectral type. The classification is performed first by merging the real data with a small sample of synthetic stars, which are representative of the solar neighbourhood stellar content and which can be used as tracers of the properties of associated real stars. Secondly, the overall data are classified - without supervision - using the Kohonen neural network algorithm, which takes all the features considered into account simultaneously. The results and advantages of this approach suggest that the finite mixture of real and synthetic stars provides a useful framework for analysing the overlapping spatial, kinematic and spectrophotometric distributions of stellar populations.