With the increasing data dimensionality, attribute reduction has become an effective way to eliminate effects caused by redundant attributes.
Due to the better performance, neighborhood rough sets (NRS) have been widely applied to attribute reduction.
It uses a fixed neighborhood radius to calculate the neighborhood relation between each object.
However, the neighborhood radius of NRS is influenced by the data distribution and needs to be optimized by grid searching.
This causes performance degradation, leading to a time complexity of not lower than O(N2).
Motivated by this, we propose the hypersphere neighborhood rough sets method (HNRS), a novel NRS method with time complexity O(N).
This method maps the data to a new feature space and builds a set of hyperspheres, adaptively generating a different neighborhood for objects in each class.
Extensive experiments on different datasets consistently demonstrate that our proposed method significantly outperforms state-of-the-art baselines.
自己的东西太少。提炼出:三个技术或者三个阶段,或者三个技术及其各自优势(6句)
Neighborhood rough sets (NRS) have been sucessfully applied to attribute reduction for both symbolic and numeric data.
Popular approaches has a time complexity of not lower than O(MN^2).
In this paper, we propose a hypersphere neighborhood rough sets (HNRS) algorithm with time complexity O(MN).
First, the data is mapped to a new feature space xxx.
Second, a set of hyperspheres is built xxx.
Third, neighorhood of objects are generated according to both class label and instance data.
feature extraction or attribute reduction?