Ridge-SimSel: A generalization of the variable selection method SimSel to multicollinear data sets

Variable selection is an important part of most applications of statistical modeling. The variable selection problem is in biological applications often demanding, due to that the true underlying function in general is nonlinear and that datasets often contain multicollinearities, measurement-error variables, and outliers. We here introduce Ridge-SimSel, a simulation based variable selection method that is quite insensitive with respect to these complexities. Ridge-SimSel is a generalization of the previously described SimSel method to handle the case of multicollinear datasets. It works by fitting an approximative model to the data, which is used to study how disturbances of the independent variables affect the fit of the approximative model. The main idea is that disturbing unimportant variables does not affect quality of the model fit.