Genetic Algorithm Based Variable Selection for Partial Least Squares Regression Using ICOMP Criterion

Partial least squares regression is a statistical method of modeling relationships between YNxM response variable and XNxK explanatory variables which is particularly well suited for analyzing when explanatory variables are highly correlated. In partial least square part, some model selection criteria are used to obtain the latent variables which are the most relevant variables describing the response variables. In this study, we investigate the performance of Partial Least Squares Regression-the Nonlinear Iterative Partial Least Squares (PLSR-NIPALS), Partial Least Squares Regression-the Variable Importance in the Projection (PLSR-VIP) and the Genetic Algorithms Partial Least Square Regression (GAPLSR) when the fitness function is the Information Complexity Criterion (ICOMP) for model selection. We compared the performance of these methods with real world data and simulation data sets and used the adjusted R square (R2adj) values to quantify the adequacy of the models.