A novel approach for efficient updating approximations in dynamic ordered information systems

• Published in: Information sciences 2020-01, Vol.507, p.197-219
• Main Authors: Wang, Shu, Li, Tianrui, Luo, Chuan, Hu, Jie, Fujita, Hamido, Huang, Tianqiang
• Format: Article
• Language: English
• Subjects: Dominance-based rough sets; Knowledge discovery; Ordered information system; Rough sets
• ISSN: 0020-0255; 1872-6291
• DOI: 10.1016/j.ins.2019.08.046

Dynamic data in real-time application are typically updating in a multi-dimensional manner. In this paper, we introduce a novel approach based on Dominance-based Rough Set Approach (DRSA) to efficiently deal with the multi-dimensional variations of attribute set and attribute values in dynamic Ordered Information Systems (OIS). We improve the original notion of the P-generalized decision domains to make the feature value matrix be dominance symmetrical, and propose an efficient strategy based on the improved notion to obtain the dominance feature matrix. Then, we employ the dominance-feature-matrix-based incremental strategy to avoid repeated comparisons between original attributes, so that to efficiently update rough approximations of DRSA with the simultaneously increased attribute set and varied attribute values. In our approach, the steps based on these two combined strategies can work altogether or separately, not only efficiently dealing with the simultaneously increased attribute set and varied attribute values, but also efficiently dealing with the individually increased attribute set or varied attribute values in dynamic OIS. Efficient algorithm based on the updating strategies is designed and multiple groups of experiments are conducted. Experimental results on different real-world data sets show that the proposed algorithm is much faster than other algorithms for dealing with the multi-dimensional or the single-dimensional variations of attribute set and attribute values.