摘要
差别矩阵非空元素的个数,直接影响基于差别矩阵的Rough集属性约简算法的效率。分析了几种差别矩阵的不足,基于此,重新定义了一种差别矩阵,该差别矩阵把划分U/C={[x1]C,[x2]C,…,[xn]C}的一个等价类看成一条规则参与区分,从而大大减少了差别矩阵非空元素的个数,提高了Rough集属性约简算法的效率。给出了这几种差别矩阵非空元素的计算公式及其相关定理。提出了一种带启发式知识的约简算法,该算法在很大程度上能找到决策表的最小属性约简。最后给出了对UCI一些数据库的仿真结果。
The efficiency of algorithm for attribute reduction in rough set theory based on discernibility matrix was impacted by the number of non-empty elements in discernibility matrix. The disadvantages of some discernibility matrices were analyzed. According to this, a new discernibility matrix was redefined, which regarded a decision-making class [xj]C as a decision-making rule, where [xj]C ∈U/C. Therefore, it decreased greatly the number of non-empty elements, which improved the efficiency of algorithm for attribute reduction based on discernibility matrix. And the formulas computing the number of non-empty elements in these discernibility matrices and some theorems were introduced. An algorithm based on heuristic information was proposed. At most time, this algorithm can find out a minimal attribute reduction. Lastly, the simulation experiments for UCI database were displayed.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2008年第14期3717-3720,3725,共5页
Journal of System Simulation
基金
成都信息工程学院科研基金(CRF200719).
关键词
粗糙集
最小属性约简
差别矩阵
属性约简
rough set
minimal attribute reduction
discernibility matrix
attribute reduction