Fuzzy-rough instance selection

Abstract

Rough set theory provides a useful mathematical foundation for developing automated computational systems that can help understand and make use of imperfect knowledge. Since its introduction, this theory has been successfully utilised to devise mathematically sound and often, computationally efficient techniques for addressing problems such as hidden pattern discovery from data, feature selection and decision rule generation. Fuzzy-rough set theory improves upon this by enabling uncertainty and vagueness to be modeled more effectively. Recently, the value of fuzzy-rough sets for feature selection and rule induction has been established. However, the potential of this theory for instance selection has not been investigated at all. This paper proposes three novel methods for instance selection based on fuzzy-rough sets. The initial experimentation demonstrates that the methods can significantly reduce the number of instances whilst maintaining high classification accuracies.