水平划分决策表的属性约简算法

差别矩阵属性约简是粗糙集重要约简方法之一，但在处理不一致大数据集时存在不足。为此，提出了决策差别矩阵的概念，并给出基于决策差别矩阵的属性约简定义，同时研究了由该定义获得的约简与正区域约简之间的等价性。为了提高求解效率，给出水平划分决策表的方法，指出将划分的子决策表分配到不同的网络节点上，基于子决策差别矩阵可并行完成核属性和属性约简；并设计了并行约简算法。实例分析和UCI中数据集的实验比较表明所提出的约简算法是正确的、高效的。

An Algorithm for Attribute Reduction Based on Horizontally Partitioning Decision Table

The attribute reduction based on discernibility matrix is one of important research issues in rough set theory, which exist some shortcomings when dealing with inconsistent decision table and big data sets. So, the notion of decision discernibility matrix and definition of attribute reduction based on decision discernibility matrix are present, and it is proved that attribute reduction acquired from the definition is equivalence to attribute reduction based on positive region. For improving efficiency of reduction, this paper proposes the method of horizontally partitioning decision table, and points out that the sub-decision table can be assigned to different network nodes and finish computing core attribute and attribute reduction based on sub-decision discernibility matrix. Moreover, a parallel reduction algorithm is designed. Finally, the example analysis experiment results form datasets of UCI show that the proposed parallel algorithms are efficient and effective.