Enhancing Incremental Feature Subset Selection in High-Dimensional Databases by Adding a Backward Step

Abstract

Feature subset selection has become an expensive process due to the relatively recent appearance of high-dimensional databases. Thus, the need has arisen not only for reducing the dimensionality of these datasets, but also for doing it in an efficient way. We propose the design of a new backward search which performs better than other state-of-the-art algorithms in terms of size of the selected subsets and in the number of evaluations, by removing attributes given a smart decremental approach and, besides, it is guided using a heuristic which reduces the needed number of evaluations commonly expected from a backward search.