AN EFFICIENT FILTER-WRAPPER ATTRIBUTE REDUCTION ALGORITHM BASED ON SOFT PRUNING AND GENERALIZED WEIGHTED NEIGHBORHOOD ROUGH SETS

Abstract

Attribute reduction is a central task in rough set theory, aiming to eliminate redundant attributes while preserving the classification ability of the decision system. Within the framework of generalized weighted neighborhood rough sets (GWNRS), the existing heuristic algorithm has shown some effectiveness but still incurs a high computational cost when applied to large-scale data. To overcome this limitation, we propose a new algorithm, SP-GWNRS, based on a soft pruning mechanism. Each attribute is associated with a counter that tracks the number of times it contributes negatively to the dependency degree, enabling the algorithm to adaptively balance exploration and pruning to reduce redundant evaluations. Furthermore, a wrapper phase is incorporated to select the optimal final reduct based on classification accuracy. Experimental results on benchmark decision systems demonstrate that SP-GWNRS achieves reductions of comparable size to those of the original heuristic algorithm, while outperforming it in classification accuracy.