一种Hu差别矩阵属性约简的高效算法

• 作 者: 黄国顺 曾凡智 陈广义 文翰
• 机 构: 佛山科学技术学院理学院,机电与信息工程学院,
• 出 处: 华中科技大学学报(自然科学版)
• 发表日期: 2012年04期
• 关键词: 差别矩阵 Hu Hu差别矩阵 属性约简 相对知识量 决策表 属性 递归计算 算法 属性重要性

针对决策表,首先从理论上证明了相对知识量与新条件信息熵的等价性,从而保证可以从相对知识量的角度计算Hu差别矩阵属性约简.为快速计算出决策表属性的相对知识量,给出一种相对知识量的递归计算公式,从而得到新条件信息熵的递归计算公式.以此为基础给出一种属性重要性定义并设计出一种计算Hu差别矩阵属性约简的启发式算法.算例和实验结果验证该算法是高效的.

For the decision table, the equivalence between the relative knowledge and the new conditional information entropy is theoretically proved, so as to ensure that the Hu difference matrix attribute reduction can be calculated from the perspective of relative knowledge.For the purpose of quickly calculating the relative knowledge of decision table attributes , A recursive formula of relative knowledge is given and a recursive formula for calculating the entropy of new condition information is obtained.On the basis of this, a definition of the importance of attributes is given and a method to calculate the attribute reduction of Hu difference matrices is designed Heuristic Algorithm. The examples and experimental results verify that the algorithm is efficient.