摘要
针对目前决策表属性约简的计算问题,研究了粗糙集理论中差别矩阵,讨论了知识粒度与信息量、类别特征矩阵之间的关系,利用知识粒度最大的属性生成较小的类别特征矩阵,设计了新的启发式规则来快速缩小搜索空间和最小化属性选择,提出了一个基于知识粒度的最小属性约简算法,并用一个实例证明了算法的正确性。与类别特征矩阵相比,采用知识粒度生成的类别特征矩阵可以有效地减少存储空间。实验结果表明,所提出的算法能够得到最小属性约简。
The problem of calculating the attribute reduction of a decision table is studied. By the research of discernibility matrices in rough sets, the improved elass feature matrices were presented. The relationships among knowledge granulation, information quantity and class feature matrices were discussed. The new heuristic rules for reducing the search spaee and minimizing the selecting attribute sets were designed. Based on that, a minimal attribute reduction algorithm based on the knowledge granulation was proposed and the correctness of this algorithm was proved with an example. Compared with the algorithms based on class feature matrices, this algorithm is of much less space complexity and time complexity. The experiment results show that the minimal attribute reduction can be got.
出处
《江苏理工学院学报》
2008年第2期22-26,共5页
Journal of Jiangsu University of Technology
基金
江苏技术师范学院青年基金项目(kyy07030)
关键词
粗糙集
最小属性约简
知识粒度
不一致决策表
rough set
minimax attribute reduction
knowledge granulation
inconsistent decision table
