Bootstrap Information Criterion for Selection of Variables in Canonical Correlation Analysis
Yasunori Fujikoshi, Tetsuro Sakurai, Shingo Kanda, Takakazu Sugiyama · 2009
In this paper, we consider the problem for selection of variables based on selection of redundancy models in canonical correlation analysis. Under normality, DIC and CDIC were proposed by Fujikoshi [2] and Fujikoshi and Kurata [3]. Our purpose is to extend such approaches to nonnormality and to propose bootstrap information criterion for selection of redundancy models. By some simulations it is shown that our bootstrap information criterion is useful in nonnormal situations.