Related families-based methods for updating reducts under dynamic object sets

Due to rapid growth of data with respect to time, feature selection in dynamic covering decision information systems (DCDISs) is an important research direction of covering rough set theory, and we have not observed researches on related families-based methods for updating reducts of DCDISs with dynamic object variations. In this paper, we first introduce the concepts of covering decision approximation spaces (CDASs) and dynamic covering decision approximation spaces (DCDASs) when varying object sets, and illustrate the relationship between related sets of CDASs and those of DCDASs. Then incremental learning methods based on related families are provided for feature selection in DCDISs with the addition and deletion of objects. Finally, we develop the corresponding heuristic incremental algorithms for feature selection in DCDISs and employ experimental results on benchmark datasets to demonstrate that these algorithms give satisfactory results in terms of running times.