Partial least squares method based on least absolute shrinkage and selection operator

Abstract

In many multivariate statistical techniques, a set of linear functions of the original variables is produced. But this kind of model derived is difficult to interpret, Such as principle component regression (PCR) and partial least squares regression (PLSR), they cannot select variables. The approach least absolute shrinkage and selection operator (LASSO) can easily produce sparse solutions and select variables during estimate parameters. This article proposes a new technique for interpretation based on these properties, it's a combination of partial least squares (PLS) and LASSO and can easily interpret regression models. This method will be more favorable for large number of variables compared to PLS.