Attribute reduction based on adaptive neighborhood rough sets and three-way pied kingfisher optimizer

Attribute reduction is a critical research topic in rough set theory, aiming to eliminate irrelevant and redundant attributes while maintaining the descriptive power of the data. However, traditional neighborhood rough set- based attribute reduction methods typically require manual setting of the parameters involved in the methods (e.g., neighborhood radius), and often struggle to find the globally optimal feature subset. To address these issues, this paper proposes a novel attribute reduction algorithm (called 3PKO-ANRAR) based on adaptive neighborhood rough sets and the three-way Pied Kingfisher Optimizer (PKO) algorithm. First, the neighborhood relationships are constructed using sample distribution information, and a neighborhood radius reduction factor is defined to enable reasonable adaptation of the neighborhood radius. Second, PKO is introduced as an efficient search mechanism, where the position of the kestrels is treated as the result of attribute reduction (i.e., the reduct). A fitness function is defined based on the attribute dependency of the adaptive neighborhood rough sets to evaluate the quality of the reduct. Third, to mitigate the risk of PKO getting trapped in local optima, several improvements are introduced, including the incorporation of a three-way group partitioning mechanism and a local perturbation strategy, allowing for dynamic updates of the kingfishers' positions during iteration. The proposed attribute reduction algorithm based on adaptive neighborhood rough sets and three-way PKO is able to find feature subsets with minimal information loss while achieving high classification accuracy.