Heuristic Attribute Reduction Based on Neighborhood Knowledge Granularity

The classical attribute reduction algorithm is not suitable for the neighborhood decision information system of numerical attribute. Since any feature subset in real feature space can be approximated by neighborhood information particles, the concept of knowledge granularity is extended to neighborhood rough set from the perspective of the granularity computation. In this paper, knowledge granularity, attribute importance and heuristic algorithm based on the granularity of knowledge in neighborhood rough set are studied from the point of view of neighborhood relation matrix. The feasibility of heuristic algorithm is analyzed on UCI data sets. Experiments achieve better classification accuracy and lower attribute reduction results than the existing algorithms.