基于区分能力的HU差别矩阵属性约简算法

将决策表属性区分能力大小与HU差别矩阵结合起来,建立起属性区别能力与该属性在HU差别矩阵中出现次数之间的关系.提出与HU差别矩阵属性约简相适应,只依赖于等价类基数计算的属性区分能力计算公式,得到HU差别矩阵约简的属性区分能力判定定理.定义了以属性区分能力为基础的属性相对重要性概念,提出以相对重要性为启发式信息的HU属性约简算法.由于该算法不必构造差别矩阵,只依赖于等价类基数的计算,从而大大提高了算法效率.数值算例和实验结果表明,该算法更有利于最优或次优约简结果的搜索.

An Algorithm of Attribute Reduction Based on HU's Discernibility Matrix by Distinguishing Capability

The relationship between HU′s discernibility matrix and attribute discernibility for decision table is established by combining with them.A formula of attribute discernibility associated with HU′′s discernibility matrix is proposed,which only depends on the computational efficiency of equivalent class.A judgement theorem of HU′s attribute reduction is presented by attribute discernibility.The relative significance is defined by attribute discernibility,then a heuristic attribute reduction algorithm based on this significance is presented for decision table.It improves the computational efficiency largely for avoidance of the constructing process of discernibility matrix.Numerical example and experimental results show that the algorithm we propose will favor the exploring of optimal or sub-optimal reduction.