Hierarchical classification model of attribute decomposition approach based on rough set

In data mining, it is difficult to construct classification model for massive high-dimension databases. Time complexity of computation is high, and obtained classification models are difficult to understand or interpret. Based on rough set theory, this paper proposed a new attribute decomposition approach to discover concept hierarchy in the database and establish hierarchical classification models. For familiar databases with prior knowledge, several attributes are grouped together; for unfamiliar databases, attributes are selected and grouped together according to data table decomposition measure; and then objects' classes are re-labeled according to the coincidence search indicator proposed in this paper. Then discover intermediate concept layer, construct hierarchical classification models, divide massive high-dimension databases into small databases hierarchically. Besides, because intermediate concept layers have certain physical meaning, the understandability of the model is greatly improved. Finally, this paper validated the effectiveness of this algorithm with cases and public datasets from UCI. The result shows that this algorithm can produce hierarchical classification models with clear hierarchies, strong understandability, while still keeping high classification rate.