Abstract:

Attribute reduction is an important part in rough set theory. For the usual decision table, when considering each row in the table as a decision rule, the number of occurrences of the same decision rule is taken as the weight, and thus the weighted decision table is obtained. The corresponding matrix of positive region reduction is proposed and the proof is given, and the algorithm 1 is obtained. Comparing the positive region reduction algorithm between the decision table and the weighted decision table, it is found that the algorithm 1 is superior to the former to some extent after converting the decision table into a weighted decision table. Then the corresponding matrix of approximate classification accuracy reduction is proposed and proved. Two algorithms are tested on the selected UCI datasets. The feasibility and effectiveness of the proposed algorithm are illustrated by experiments.

Key words: rough set, attribute reduction, weighted decision table, positive region reduction, approximate classification accuracy reduction

摘要：

属性约简是粗糙集理论的重要应用。考虑将决策表中的每行都作为一条决策规则时，若把表中出现相同决策规则的次数作为权，可得到带权决策表。提出了关于带权决策表的正域约简相应的辨识矩阵并给出了证明，从而得到了约简算法。相比于决策表中的正域约简时发现，通过将决策表转化为带权决策表后，再利用算法1进行约简时，其在一定程度上优于前者。提出了近似分类精度约简相应的辨识矩阵并给出了证明。对于2个算法，在选取的UCI数据集上进行了实验验证。通过实验进一步说明了所提出算法的可行性和有效性。

关键词: 粗糙集, 属性约简, 带权决策表, 正域约简, 近似分类精度约简