Lasso Penalized Model Selection Criteria for High-Dimensional Multivariate Linear Regression Analysis
Shota Katayama and Shinpei Imori
Abstract
This paper proposes two model selection criteria for identifying relevant predictors in
the high-dimensional multivariate linear regression analysis.
The proposed criteria are based on a Lasso type penalized likelihood function to allow the high-dimensionality.
Under the asymptotic framework that the dimension of multiple responses goes to infinity while the maximum
size of candidate models has smaller order of the sample size,
it is shown that the proposed criteria have the model selection consistency, that is, they can asymptotically pick out
the true model.
Simulation studies show that the proposed criteria outperform existing criteria when the dimension of multiple
responses are large.
Paper
http://dx.doi.org/10.1016/j.jmva.2014.08.002