A fast feature selection approach based on rough set boundary regions

Dataset dimensionality is one of the primary impediments to data analysis in areas such as pattern recognition, data mining, and decision support. A feature subset that possesses the same classification power as that of the whole feature set is expected to be found prior to performing a classification task. For this purpose, many rough set algorithms for feature selection have been developed and applied to incomplete decision systems. When they address large data, however, their undesirable efficiencies could be intolerable. This paper proposes a boundary region-based feature selection algorithm (BRFS), which has the ability to efficiently find a feature subset from a large incomplete decision system. BRFS captures an inconsistent block family to construct a rough set boundary region and designs a positive stepwise mechanism for the construction of boundary regions with respect to multiple attribute subsets, making the acquisition of boundary regions highly efficient. The boundary regions are used to build significance measures as heuristics to determine the optimal search path and establish an evaluation criterion for rules to identify feature subsets. These arrangements make BRFS capable of locating a reduct more efficiently than other available algorithms; this finding is supported by experimental results. (C) 2013 Elsevier B.V. All rights reserved,