The lasso and rlasso: A Comparative study

Abstract

Linear multiple regression models are the most widely used statistical models to illustrate the effect of a set of covariates on an outcome of interest. However, only a small number of covariates actually has an influence on the outcome of interest. The problem of choosing the true subset of covariates within a multiple linear regression model has received considerable attention over the years. In this paper, we compare the performance of two regularization approaches in this study: the least absolute shrinkage and selection operator (lasso) and the reciprocal lasso (rlasso). Simulation results show that both approaches outperform in terms of prediction accuracy. The results of both approaches (lasso and rlasso) are very similar. Our outcomes demonstrate that lasso and rlasso perform comparably in various simulation studies.