Feature selection for regression problems can be highly beneficial in terms of robustness and execution speed.

Most common filter algorithms for feature selection are focused on the interactions of the features with the output variable, giving diminished attention to the interactions among features. Therefore, their selected subset of features does not consider features that by themselves are useless but when taken with others, they provide a significant performance improvement on the regression algorithm. The aim of this project is to consider both kinds of interactions in order to obtain a more robust feature subset.