Fast algorithms for computing rough approximations in set-valued decision systems while updating criteria values

Abstract

Rough set theory is advocated as a framework for conceptualizing and analyzing various types of data, which is a powerful tool for discovering patterns in a given data set through a pair of concepts, namely, upper and lower approximations. Strategic behaviors need to be reinforced continuously under the dynamic decision environment where data in the decision process can change over time. Incremental learning is an effective technique to deal with dynamic learning tasks since it can make full use of previous knowledge. Set-valued data, in which a set of values are associated with an individual, is common in real-world data sets. Motivated by the needs of knowledge updating due to the dynamic variation of criteria values in the set-valued decision system, in this paper, we present the updating properties for dynamic maintenance of approximations when the criteria values in the set-valued decision system evolve with time. Then, two incremental algorithms for computing rough approximations are proposed corresponding to the addition and removal of criteria values, respectively. Experimental results show our incremental algorithms work successfully on datasets from UCI as well as artificial datasets, and achieve better performance than the traditional non-incremental method.