umich-cphds/snif: Selection of Nonlinear Interactions by a Forward Stepwise Algorithm

Variable selection methods specifically targeted toward interactions between predictor variables are limited. The 'snif' package provides an implementation of "Selection of Nonlinear Interactions by a Forward stepwise method" by Narisetty, Mukherjee et al (2091) <doi:10.1002/sim.8059>. SINF incorporates nonlinearity of the predictors by introducing basis function expansions of the predictors and creates a forward selection path for main and interaction effects following the strong heredity principle (ie, interactions are present only when both the corresponding main effects are present). In addition to adding the basis functions for each predictor to account for nonlinearity, SNIF retains the linear terms so that the basis functions for a predictor are used only when the linear term is not sufficient to explain its effect on the outcome.

install.packages("remotes")
remotes::install_github("umich-cphds/snif")