Database mining on derived attributes granular and rough computing approach

Abstract

Selecting a desirable set of attributes is essential in data mining. In this paper, we develop the notion of universal model, on which a complete set of all possible derived attributes can be generated. Then, for any data mining task, the relation can select desirable attributes by interpreting them from the universal model. The development of the universal model is based on the relation lattice, which was initiated by T. T. Lee around 1983. However, we define the lattice differently. The relation lattice theory, as rough sets, is based on the partitions induced by attributes.