An incremental approach to feature selection using the weighted dominance-based neighborhood rough sets

Dominance-based neighborhood rough set (DNRS) is capable to give qualitative and quantitative descriptions of the relations between ordered objects. In spite of its effectiveness in feature selection, DNRS ignores the various significance of features. In fact, different features exert different impacts on decision-making. Once we explore these differences in advance, it is easier to find out features with high correlation and dependency. Likewise, it is inevitable that in big-data era the objects may update from time to time, which calls for efficient attribute reduction. However, the existing approaches are inappropriate for the weighted and ordered data. Motivated by these two deficiencies, first, we assign different weights to conditional attributes and establish the weighted dominance-based neighborhood rough set (WDNRS). Then a kind of conditional entropy in matrix form and ensuing updating principles are put forward to evaluate the significance of the attributes. In addition, grounded on the entropy, we come up with the heuristic algorithm and corresponding incremental mechanism when objects increase. Finally, twelve experiments are carried out to verify that it is effective and efficient for the designed method to select features in dynamic datasets.