Incremental reduction methods based on granular ball neighborhood rough sets and attribute grouping
Yan Li,Xiaoxue Wu,Xizhao Wang
DOI: https://doi.org/10.1016/j.ijar.2023.108974
IF: 4.452
2023-07-01
International Journal of Approximate Reasoning
Abstract:In the framework of rough sets, incremental algorithms can effectively reduce repetitive computations for dynamic datasets by discovering the update principles of relevant knowledge. Considering the variations in attribute sets, incremental methods based on neighborhood granulation provide fast updates to attribute reduction in continuous-valued information systems. However, most of these methods assume that the sample distribution and neighborhood radius remain fixed with changes in the attribute set, which may mislead the granulation process and affect the model's classification ability. In this paper, a new incremental reduction method based on granular balls and attribute grouping is proposed for dynamic information systems with multiple attribute additions. Different neighborhood radii are adaptively determined when the attribute set changes, and the number of neighborhood granules can also be effectively reduced. Furthermore, the attributes are grouped based on the k-means algorithm, and only attributes from different groups or with small relevance to those in the current reduction set are considered to be incorporated as a reduction attribute, thus reducing the computation time and simultaneously preserving informative attributes. Linear and nonlinear correlation coefficients are used to measure attribute relevance based on which three corresponding incremental reduction algorithms are developed. Finally, extensive experimental results on 12 benchmark datasets are shown to compare the proposed method with non-incremental and typical incremental methods in terms of time cost, classification accuracy, the size of attribute reduction, and coverage of the reduction.
computer science, artificial intelligence