B0269
Title: Robust and sparse estimation of Gaussian graphical models based on Winsorization
Authors: Ginette Lafit - University of Leuven (Belgium)
Francisco J Nogales - Universidad Carlos III de Madrid (Spain)
Marcelo Ruiz - Universidad Nacional de Río Cuarto (Argentina) [presenting]
Ruben Zamar - University of British Columbia (Canada)
Abstract: Robust and sparse estimation of Gaussian graphical models based on Winsorization. The use of a robust covariance estimator is proposed based on multivariate Winsorization in the context of the Tarr-Muller-Weber framework for sparse estimation of the precision matrix of a Gaussian graphical model. Likewise, with Croux-Ollerer's precision matrix estimator, the proposed estimator attains the maximum finite sample breakdown point of 0.5 under cellwise contamination. An extensive Monte Carlo simulation study is conducted to assess the performance of this and the currently existing proposals. It is found that this proposal has a competitive behaviour, regarding the estimation of the precision matrix and the recovery of the graph. The usefulness of the proposed methodology is demonstrated in a real application to breast cancer data.
