Academic Editor: Youssef EL FOUTAYENI
Received |
Accepted |
Published |
14 November 2019 |
29 November 2019 |
10 March 2020 |
Abstract: Partial least squares regression (PLS regression) is often used to overcome the problem of multicollinearity in chemometrics. This method shrinks the ordinary least squares estimator (OLS estimator) in order to exchange bias with lower variance. PLS requires a proper model selection tool to choose the optimal model; cross validation is the most common tool in the PLS literature. In this work we will compare leave one out cross validation with two methods of generalized cross validation (GCV), often used in regularization methods in the numerical analysis literature. The difference between the two (GCV) methods is the use of different estimations for the trace of the influence matrix. Our comparison is conducted using simulated datasets.
