A strategy of attribute reduction based on partition
Hui Wang1, Tao Zheng1, Weiwei Zhang2
The attribute reduction is an important pre-processing step for data mining. In order to avoid striking equivalence classes repeatedly for positive region or information entropy reduction it is proposed to calculate attribute reduction by constructing partition directly. At the same time the judgments of the absolute reduction and the relative reduction based on the equivalent division are proved. And the data description quality for the relative reduction has been defined. It is shown that striking minimum relative reduction of decision table is in the cost of the relative decline of description quality for classification.