Feature selection based on composition of rough sets induced by feature granulation

Abstract The term “feature selection” refers to the problem of selecting the most predictive features for a given outcome. Reducing the number of features is important to lower the computation cost of algorithms and also to achieve better generalization capabilities. The rough set theory has rapidly established itself as an effective tool for finding the smallest sets of features without any loss of information. This paper introduces a new approach based on the idea of exploiting rough set theory to compose partitions of the data induced by feature granulation. The partitions are iteratively aggregated in a single representative partition and, at each iteration, the obtained partition is used to evaluate the quality of the subset of features selected, thus reducing the cost of each evaluation. The consequence of using rough set theory is an algorithm that presents good reduction capability with less computational complexity w.r.t. Quickreduct algorithm. The proposed approach, called Roughinement, has been compared with approaches recently appeared in literature yielding comparable to better results.