“Modified LASS Method Suggestion as an additional Penalty on Principal Components Estimation – with Application-“

Abstract

    This research deals with a shrinking method concernes with the principal components similar to that one which used in the multiple regression “Least Absolute Shrinkage and Selection: LASSâ€. The goal here is to make an uncorrelated linear combinations from only a subset of explanatory variables that may have a multicollinearity problem instead taking the whole number say, (K)  of them. This shrinkage will force some coefficients to equal zero, after making some restriction on them by some "tuning parameter" say, (t) which balances the bias and variance amount from side, and doesn't  exceed the acceptable percent explained variance of these components. This had been shown  by MSE criterion in the regression case and the percent explained variance in the principal components case.