一种基于分明矩阵的启发式知识约简方法

提出了基于分明矩阵的启发式知识约简方法。在决策表的相对约简过程中采用分明矩阵来表达知识,并利用分明矩阵中项的长度和每个属性的频率作为启发信息进行属性的选择。现已证明,寻找决策表中最小相对约简问题是典型的NP hard问题。所提供的算法在大多数情况下能够找到最小约简,即使在没找到最小约简的情况下,也能找到次优解。通过实例分析,证明该算法是求解属性相对约简的快速有效的方法

Approach to heuristic knowledge reduction based on discernibility matrix

An approach to heuristic knowledge reduction based on discernibility matrix is proposed. Knowledge is expressed first using discernibility matrix during the decision table's relative reduction and attributes are selected according to the elements' length in the discernibility matrix and the frequency of each attribute. It has been proved that the problem of searching minimum relative reduction is an NP hard problem. In the most cases the minimum reduction can be obtained using the proposed method. Otherwise, there must be a feasible solution. The practical results show that the approach is quick and effective in solving relative reduction problem.