Flow-based tolerance rough sets for pattern classification

Rough set theory is a useful mathematical tool for pattern classification to deal with vagueness in available information. The main disadvantage of rough set theory is that it cannot handle continuous attributes. Although various discretization methods have been proposed to deal with this problem, discretization can result in information loss. It has been found that tolerance rough sets with a tolerance relation can operate effectively on continuous attributes. A tolerance relation is related to a similarity measure which is commonly defined by a simple distance function to measure the proximity of any two patterns distributed in feature space. However, for a simple distance measure, it oversimplifies the criteria aggregation resulting from not considering attribute weights, and it is not a unique way of expressing the preference information on each attribute for any two patterns. This paper proposes a flow-based tolerance rough set using flow, which represents the intensity of preference for one pattern over another, to measure similarity between two patterns. To yield high classification performance, a genetic-algorithm based learning algorithm has been designed to determine parameter specifications and generate the tolerance class of a pattern. The proposed method has been tested on several real-world data sets. Its classification performance is comparable to that of other rough-set-based methods. (C) 2014 Elsevier B.V. All rights reserved.