On the robustness of the cosine distribution depth classifier

Investigating how classifiers perform under some data contaminations is an important issue in robustness studies. While some research is available on the robustness of classifiers, a little is known about directional classifiers. This work thus investigates the robustness of the cosine depth distribution classifier, a classification technique recently introduced for directional data. This latter is a non-parametric method and it is based on the distribution function of the cosine depth.