Quasi-Relief Method of Informative Features Selection for Classification
С. А. Субботин · 2018 IEEE 13th International Scientific and Technical Conference on Computer Sciences and Information Technologies (CSIT) · 2018
The feature selection problem for classification is addressed. The modified Relief method is proposed. It selects a subsample from the original sample, computes hashes for instances, used to select and delete redundant instances, compute feature similarity and difference, and updates the feature weights. The experiments to study the proposed method have been carried out. They showed that the proposed method provides a significant acceleration of computations.