基于可区分度的连续空间属性约简算法研究.

Aiming at the problems of information loss in continuous data processing, inconsistent information introduced by granulation strategies, and difficulty in parameter optimization in the existing rough set attribute reduction methods, this paper proposed a non-monotonic heuristic attribute reduction algorithm based on category discrimination which was suitable for continuous data. Firstly, it divided the universe according to the label of each sample, and combined the samples of the same label into a cluster, and defined the inter-class discrimination and intra-class discrimination of each cluster. Secondly, it defined a new attribute importance criterion to determine the optimal reduction set, so as to improve the classification performance of subsequent classifiers. The proposed algorithm was compared with other 6 attribute reduction algorithms on the 11 UCI datasets. The results show that compared with the 6 algorithms, the average dimension of the reduction sets obtained by the proposed algorithm is reduced by 1. 16, and the average classification accuracy is improved by 3. 42%, showing better reduction performance. [ABSTRACT FROM AUTHOR]