A0761
Title: Optimization methods for best subset selection problem in high-dimensional linear dimension reduction models
Authors: Sarat Moka - The University of New South Wales (Australia) [presenting]
Zdravko Botev - The University of New South Wales (Australia)
Benoit Liquet - Macquarie University (Australia)
Samuel Muller - Macquarie University (Australia)
Abstract: Principal components analysis and partial least squares are two popular methods used for dimensionality reductions, and they have numerous applications in several fields, including economics. Both these methods build principal components, which are new variables that are combinations of all the original variables. A key drawback of these principal components is the difficulty of interpreting them when the number of variables is large. To overcome this difficulty, it is desirable to define principal components from the most relevant variables. However, selecting the most relevant variables is an NP-hard problem, particularly challenging in high-dimensional settings where the number of features can be far higher than the number of observations. The problem is stated as a best subset selection problem, and new efficient optimization methods have been developed to address this problem. Several empirical experiments are provided to illustrate the efficacy of the approaches.
