A novel model of fuzzy rough sets based on grouping functions and its application

The grouping and overlap functions play a prominent role in areas such as classification and image processing. On the one hand, the grouping function, as an aggregation function closely related to the t-conorm, has not been specifically used to build fuzzy rough set (FRS). Thus, a novel FRS model based on the grouping function is proposed. On the other hand, in the context of big data, directly analysing all the attributes will increase the computational complexity, so attribute reduction (AR) is necessary. The upper approximation contains boundary region and lower approximation informations, which has certain advantages. However, it is rare to specifically consider reduction from upper approximation. Therefore, the grouping functions and fuzzy negations to determine fuzzy rough set (GNFRS) reduction algorithm was designed, which utilises the advantages of the upper approximation. Finally, the GNFRS reduction algorithm is verified to have the same or higher classification accuracy compared to some other existing reduction algorithms by conducting 450 experiments on 15 public datasets.