Principal Component Regression by Principal Component Selection

Hosung Lee,Yun Mi Park,Seokho Lee
DOI: https://doi.org/10.5351/csam.2015.22.2.173
2015-03-31
Communications for Statistical Applications and Methods
Abstract:We propose a selection procedure of principal components in principal component regression. Our method selects principal components using variable selection procedures instead of a small subset of major principal components in principal component regression. Our procedure consists of two steps to improve estimation and prediction. First, we reduce the number of principal components using the conventional principal component regression to yield the set of candidate principal components and then select principal components among the candidate set using sparse regression techniques. The performance of our proposals is demonstrated numerically and compared with the typical dimension reduction approaches (including principal component regression and partial least square regression) using synthetic and real datasets.
What problem does this paper attempt to address?