Discriminant Analysis for the von Mises-Fisher distribution
Résumé
The von Mises-Fisher distribution is widely used for modeling directional data. In this paper we derive the discriminant rules based on this distribution to assign objects into pre-existing classes. We determine a distance between two von Mises-Fisher populations and we calculate estimates of the misclassification probabilities. We also analyse the behavior of the distance between two von Mises-Fisher populations and of the estimates of the misclassification probabilities when we modify the parameters of the populations or the samples size or the dimension of the sphere. Finally, we present an example with real spherical data available in the literature.
Origine : Fichiers produits par l'(les) auteur(s)