Outlier detection is an important problem in statistics that has been addressed in a variety of research areas and applications domains. In this paper, we tackle this problem using robust principal component analysis. We consider different robust estimators along with the classical estimator of principal components and develop a simulation study to compare the envisage outlier detection methods in two different scenarios: semi-supervised, where we have a training set composed only by regular observations, and an unsupervised scenario, where nothing is known about the class (regular or outlier) of each training observation

CEMAT - Center for Computational and Stochastic Mathematics