Related families-based attribute reduction of dynamic covering decision information systems

Many efforts have focused on studying techniques for selecting most informative features from data sets. Especially, the related family-based approaches have been provided for attribute reduction of covering information systems. However, the existing related family-based methods have to recompute reducts for dynamic covering decision information systems. In this paper, firstly, we investigate the mechanisms of updating the related families and attribute reducts by the utilization of previously learned results in dynamic covering decision information systems with variations of attributes. Then, we design incremental algorithms for attribute reduction of dynamic covering decision information systems in terms of attribute arriving and leaving using the related families and employ examples to demonstrate that how to update attribute reducts with the proposed algorithms. Finally, experimental comparisons with the non-incremental algorithms on UCI data sets illustrate that the proposed incremental algorithms are feasible and efficient to conduct attribute reduction of dynamic covering decision information systems with immigration and emigration of attributes.